Two gladiators enter the ring and only one leaves.
I’m sure that you’ve heard or seen that famous line from time to time.
This handy trope implies that there can only be one winner. You either win or you lose. The winner survives and continues onward. The loser, well, we tend to assume they are left in the dust.
Some have been wondering whether this perhaps loosely describes the otherwise momentous event this week entailing the long-awaited and immensely anticipated unveiling and formal release of a generative AI app known as GPT-4.
You see, a lot of hype preceded the GPT-4 release. A lot of hype. Huge quantities. Gargantuan hype.
GPT-4 is essentially the successor to the widely and wildly popular ChatGPT. Both are products by AI maker OpenAI.
As I covered in my column coverage about the GPT-4 launch, see the link here, GPT-4 provides quite notable advances and improvements over ChatGPT. Pundits would generally agree that GPT-4 seems to be more fluent in its generative AI capabilities, appears to be faster, and has thankfully seemed to be architected to reduce though certainly not eliminate the otherwise frequent odds of generating factual errors, falsehoods, and so-called AI hallucinations (I don’t like the term “AI hallucinations” since it is a form of anthropomorphism, see my discussion at the link here, but it has caught on and we seem to be stuck with it).
In the months leading up to the GPT-4 release, the CEO of OpenAI Sam Altman repeatedly voiced his concern that the outsized hype was beyond the pale. All manner of zany speculation was being proclaimed in the media about what GPT-4 would contain. The belief was that GPT-4 would supposedly make our heads spin and be utterly revolutionary in showcasing the far reaches of what AI can accomplish.
Essentially, some outspoken outreaches of social media pushed a ludicrous agenda that any other AI maker might as well shut their doors and close their research labs because GPT-4 was going to be the ultimate cat’s meow. Perhaps, the media influencers fervently whispered, GPT-4 will embody Artificial General Intelligence (AGI), namely being sentient or fully human-like.
You might vaguely be aware that the CEO of OpenAI, Sam Altman, said this in an interview posted on YouTube (dated January 17, 2023): “The GPT-4 rumor mill is a ridiculous thing. I don’t know where it all comes from. People are begging to be disappointed and they will be. The hype is just like… We don’t have an actual AGI and that’s sort of what’s expected of us.”
You can perhaps sympathize with the OpenAI CEO. The rampant and unexpected success of ChatGPT had made the interest in its successor akin to the likes of a blockbuster movie. Having gotten lucky with the rollout of ChatGPT, a boon far beyond anyone’s expectations at the time, the world awaited the sequel with bated breath. Rumors were floating around that GPT-4 would knock our socks off and change the world as we know it.
Any CEO would usually welcome a tsunami of free publicity leading up to their product launch. My gosh, it is so hard to break through today’s news glut to get your product or service into the public eye. There are a zillion channels of news and a gazillion pitches for new products and services. In the case of GPT-4, there was a constant drumbeat of stay-tuned as to how society will be indubitably altered as a result of this vaunted successor to ChatGPT.
Even if you know your product is darned good and a cut above your prior product, trying to strive toward a hyped reality is something no CEO can delicately cope with. When ChatGPT came out, the expectations were quite low if not altogether nonexistent. ChatGPT ultimately far exceeded those initial expectations. On a scale of 1 to 10, ChatGPT was quietly beforehand somewhere near a 1 and after release was rushed into the stratosphere nearing an unbelievable and totally unforeseen 9 or 10. For my coverage of ChatGPT as a brand and a marketing bonanza, see the link here.
How can your successor product compete in the expectations gambit when compared to the surprising blockbuster status of your prior product?
The answer is that it probably can’t.
The best that you can do is attempt to calm down the expectations. We all know that in life the usual rule of thumb is to try and exceed expectations. That’s typically going to produce the highest outcome. The expectations for GPT-4 were so through the roof that almost by definition the reality would be less than the expectations. Thus, it makes sense to try and dampen down overly hyped expectations in such a circumstance.
You want to somehow attain the Goldilocks of expectations.
Allow me to explain.
The hope is to leverage any heightened expectations to ensure that your new product or service will be supremely newsworthy. Meanwhile, you want to watch out for a potential blowback if the rampant expectations are not met. People will say that the new product just didn’t cut the mustard. This can be exasperating when the new product is nonetheless stellar. It is only getting dinged because it didn’t meet some wide-eyed fictionalized high bar that no one could have ever met.
So, you try to do a gentle pushback at the sky-high expectations. You want to walk a fine line. Don’t push too hard to lower the bar cause people will think you’ve got a disappointment in the making. Also, you do want the eagerness to carry into the unveiling, thus, you have to keep the buzz alive. The soup has to be just the right temperature, not too hot and not too cold.
We now know the reality.
The reality is that GPT-4 seems to be more capable as a generative AI application and we ought to acknowledge and applaud the accomplishments thereof. That being said, the actual capabilities have not lived up to the hype. One doubts that there was ever any chance of doing so, not even the slimmest of chances that the real GPT-4 could match the hyped imagined farfetched GPT-4.
Into this comes those that are shocked and dismayed that GPT-4 hasn’t gotten as much attention as it presumably deserves. Sure, the news did cover the launch. But not in the over-the-top manner that many thought would occur. To the chagrin of some, the feeling is that the rest of the world just doesn’t realize how big a deal GPT-4 really is.
In today’s column, I address the key reasons that GPT-4 hasn’t taken the world by storm and become the dominant and all-encompassing mania that ChatGPT has exhibited. You would of course be hard-pressed to shed a tear over the coverage of GPT-4. GPT-4 has gotten a ton of press. No doubt about it. The essence here though is regarding those that believe the successor ought to be getting as much rabid focus as has ChatGPT. They ardently believe that this second blockbuster is a blockbuster and deserves the same immense accolades as the predecessor blockbuster ChatGPT.
Let’s unpack what is taking place.
Some of you might be wondering what in fact generative AI is. Let’s first cover the fundamentals of generative AI and then we can take a close look at the pressing matter at hand.
Into all of this comes a slew of AI Ethics and AI Law considerations.
Please be aware that there are ongoing efforts to imbue Ethical AI principles into the development and fielding of AI apps. A growing contingent of concerned and erstwhile AI ethicists are trying to ensure that efforts to devise and adopt AI takes into account a view of doing AI For Good and averting AI For Bad. Likewise, there are proposed new AI laws that are being bandied around as potential solutions to keep AI endeavors from going amok on human rights and the like. For my ongoing and extensive coverage of AI Ethics and AI Law, see the link here and the link here, just to name a few.
The development and promulgation of Ethical AI precepts are being pursued to hopefully prevent society from falling into a myriad of AI-inducing traps. For my coverage of the UN AI Ethics principles as devised and supported by nearly 200 countries via the efforts of UNESCO, see the link here. In a similar vein, new AI laws are being explored to try and keep AI on an even keel. One of the latest takes consists of a set of proposed AI Bill of Rights that the U.S. White House recently released to identify human rights in an age of AI, see the link here. It takes a village to keep AI and AI developers on a rightful path and deter the purposeful or accidental underhanded efforts that might undercut society.
I’ll be interweaving AI Ethics and AI Law related considerations into this discussion.
Fundamentals Of Generative AI
The most widely known instance of generative AI is represented by an AI app named ChatGPT. ChatGPT sprung into the public consciousness back in November when it was released by the AI research firm OpenAI. Ever since ChatGPT has garnered outsized headlines and astonishingly exceeded its allotted fifteen minutes of fame.
I’m guessing you’ve probably heard of ChatGPT or maybe even know someone that has used it.
ChatGPT is considered a generative AI application because it takes as input some text from a user and then generates or produces an output that consists of an essay. The AI is a text-to-text generator, though I describe the AI as being a text-to-essay generator since that more readily clarifies what it is commonly used for. You can use generative AI to compose lengthy compositions or you can get it to proffer rather short pithy comments. It’s all at your bidding.
All you need to do is enter a prompt and the AI app will generate for you an essay that attempts to respond to your prompt. The composed text will seem as though the essay was written by the human hand and mind. If you were to enter a prompt that said “Tell me about Abraham Lincoln” the generative AI will provide you with an essay about Lincoln. There are other modes of generative AI, such as text-to-art and text-to-video. I’ll be focusing herein on the text-to-text variation.
Your first thought might be that this generative capability does not seem like such a big deal in terms of producing essays. You can easily do an online search of the Internet and readily find tons and tons of essays about President Lincoln. The kicker in the case of generative AI is that the generated essay is relatively unique and provides an original composition rather than a copycat. If you were to try and find the AI-produced essay online someplace, you would be unlikely to discover it.
Generative AI is pre-trained and makes use of a complex mathematical and computational formulation that has been set up by examining patterns in written words and stories across the web. As a result of examining thousands and millions of written passages, the AI can spew out new essays and stories that are a mishmash of what was found. By adding in various probabilistic functionality, the resulting text is pretty much unique in comparison to what has been used in the training set.
There are numerous concerns about generative AI.
One crucial downside is that the essays produced by a generative-based AI app can have various falsehoods embedded, including manifestly untrue facts, facts that are misleadingly portrayed, and apparent facts that are entirely fabricated. Those fabricated aspects are often referred to as a form of AI hallucinations, a catchphrase that I disfavor but lamentedly seems to be gaining popular traction anyway (for my detailed explanation about why this is lousy and unsuitable terminology, see my coverage at the link here).
Another concern is that humans can readily take credit for a generative AI-produced essay, despite not having composed the essay themselves. You might have heard that teachers and schools are quite concerned about the emergence of generative AI apps. Students can potentially use generative AI to write their assigned essays. If a student claims that an essay was written by their own hand, there is little chance of the teacher being able to discern whether it was instead forged by generative AI. For my analysis of this student and teacher confounding facet, see my coverage at the link here and the link here.
There have been some zany outsized claims on social media about Generative AI asserting that this latest version of AI is in fact sentient AI (nope, they are wrong!). Those in AI Ethics and AI Law are notably worried about this burgeoning trend of outstretched claims. You might politely say that some people are overstating what today’s AI can do. They assume that AI has capabilities that we haven’t yet been able to achieve. That’s unfortunate. Worse still, they can allow themselves and others to get into dire situations because of an assumption that the AI will be sentient or human-like in being able to take action.
Do not anthropomorphize AI.
Doing so will get you caught in a sticky and dour reliance trap of expecting the AI to do things it is unable to perform. With that being said, the latest in generative AI is relatively impressive for what it can do. Be aware though that there are significant limitations that you ought to continually keep in mind when using any generative AI app.
One final forewarning for now.
Whatever you see or read in a generative AI response that seems to be conveyed as purely factual (dates, places, people, etc.), make sure to remain skeptical and be willing to double-check what you see.
Yes, dates can be concocted, places can be made up, and elements that we usually expect to be above reproach are all subject to suspicions. Do not believe what you read and keep a skeptical eye when examining any generative AI essays or outputs. If a generative AI app tells you that Abraham Lincoln flew around the country in his private jet, you would undoubtedly know that this is malarky. Unfortunately, some people might not realize that jets weren’t around in his day, or they might know but fail to notice that the essay makes this brazen and outrageously false claim.
A strong dose of healthy skepticism and a persistent mindset of disbelief will be your best asset when using generative AI. Also, be wary of potential privacy intrusions and the loss of data confidentiality, see my discussion at the link here.
We are ready to move into the next stage of this elucidation.
GPT-4 Post-Launch And What Has Been Happening
Let’s now dive into the ChatGPT successor, GPT-4.
Here are the main topics that I’d like to cover with you today:
- 1) The Reality Of What GPT-4 Is
- 2) Beloved Feelings Toward ChatGPT Are Hard To Reinvent
- 3) Tech-Awkward Naming Of GPT-4 Is A Big Catchiness Problem
- 4) Complications Of Ready Availability For GPT-4 Is An Added Complication
- 5) Cumbersome Compliment That Microsoft Bing Uses GPT-4
- 6) Higher Cost Of API Use For GPT-4 Is A Challenge
- 7) Being Able To Pass Examines Is Not Especially A Public Windfall
- 8) Multi-Modal Requires Delayed Gratification Right Now
- 9) Other
I will cover each of these important topics and proffer insightful considerations that we all ought to be mindfully mulling over. Each of these topics is an integral part of a larger puzzle. You can’t look at just one piece. Nor can you look at any piece in isolation from the other pieces.
This is an intricate mosaic and the whole puzzle has to be given proper harmonious consideration.
The Reality Of What GPT-4 Is
It seems that just about everyone vaguely knows something or another about ChatGPT (which is generally based on GPT-3.5). People were quite curious regarding the degree to which GPT-4, the ChatGPT successor, would compare on a head-to-head functionality basis.
Here’s what the official OpenAI blog posting about GPT-4 states:
- “In a casual conversation, the distinction between GPT-3.5 and GPT-4 can be subtle. The difference comes out when the complexity of the task reaches a sufficient threshold — GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”
The essence of that depiction is that you might not especially notice at first glance that GPT-4 is ostensibly more fluent, faster, and more AI-capable than ChatGPT. You might have to hunt for it to see the differences. You might via happenstance stumble into the differences. Otherwise, on ordinary day-to-day usage, you would generally have a tough time discerning whether you are using ChatGPT versus using GPT-4 (albeit notwithstanding the user interface differences).
That is somewhat of a problem when you are trying to get people excited about a new product.
When a new car comes out from a major carmaker, the odds are that the new features will be fully delineated and people will flock to see the difference between the prior model and the newest model. Gosh, the snazzy new dashboard is enthralling. Wow, the addition of heated bucket seats with an automatic seatbelt fastener is fascinating. Those sleek shapes of the front headlights that can pivot are quite a showstopper.
If the new features are nearly indistinguishable from the prior model, this is a devil of a situation to market to the public at large. How do you get people excited about under-the-hood advances? Our car goes from 0 to 60 mph in 3.4 seconds, while the prior model took 3.7 seconds. Few of the public will notice or care about the hidden ingenious changes made far within the guts of the vehicle.
In a sense, the same or similar predicament confronts GPT-4.
It isn’t easy or simple to point to substantive differences between ChatGPT and GPT-4. There are subtleties involved. Unfortunately, selling subtleties can be an uphill battle. Selling outright obvious and demonstrable differences is much easier (in a little while herein, I’ll be mentioning the multi-modal capability as a considered distinctive feature, which has some complications worthy of reviewing).
Returning to the hype that preceded GPT-4, a common contention was that GPT-4 would be completely free of any generated factual errors, biases, falsehoods, and AI hallucinations. This would be a gob-smacking accomplishment and would finally quiet all those doubters and skeptics that say they won’t use generative AI because of the inherent output-producing maladies.
Here’s what the official OpenAI GPT-4 Technical Report says about GPT-4:
- “GPT-4 has the tendency to ‘hallucinate,’ i.e. ‘produce content that is nonsensical or untruthful in relation to certain sources.’ This tendency can be particularly harmful as models become increasingly convincing and believable, leading to an overreliance on them by users. Counterintuitively, hallucinations can become more dangerous as models become more truthful, as users build trust in the model when it provides truthful information in areas where they have some familiarity.”
Thus, AI hallucinations are still in the game. The same goes for the outputting of factual errors, falsehoods, biases, and so on.
The good news is that strenuous efforts have been made by OpenAI and seem to be ongoing to try and reduce the chances of AI hallucinations in GPT-4. Also, the claim is made that GPT-4 outdoes GPT-3.5 in terms of averting AI hallucinations, even though OpenAI makes clear that they still are going to occur.
- “On internal evaluations, GPT-4-launch scores 19 percentage points higher than our latest GPT-3.5 model at avoiding open-domain hallucinations, and 29 percentage points higher at avoiding closed-domain hallucinations” (source: OpenAI GPT-4 Technical Report).
The thing is, though the indicated improvements are laudable, they are once again more so a type of under-the-hood advancement. If GPT-4 could have magically met the hyped expectations such that no akin ailments exist at all in GPT-4, that would have been astoundingly monumental.
And readily sellable.
To clarify, and to provide a balance to this consideration, no one has somehow cured this dilemma. Indeed, for those who are looking for hard AI problems, I urge you to jump into these waters and help out. There is plenty of work to be done on such matters.
Anyway, here’s another instance of something that turns out to not be particularly sellable.
First, please know that ChatGPT was data trained on data from the Internet and was capped or locked with data through 2021. This means that ChatGPT has no data per se for 2022 and 2023 in terms of events and activities after the year 2021. People using ChatGPT are often dismayed when they discover this data limitation.
The hype about GPT-4 was that it would be more up-to-date and would include 2022, and 2023, and possibly be working in real-time to adjust and encompass the latest postings on the Internet.
Here’s what the OpenAI GPT-4 Technical Report says:
- “GPT-4 generally lacks knowledge of events that have occurred after the vast majority of its pre-training data cuts off in September 2021 and does not learn from its experience. It can sometimes make simple reasoning errors which do not seem to comport with competence across so many domains, or be overly gullible in accepting obviously false statements from a user.”
Ergo, GPT-4 is also landlocked with data stuck in time.
I could go on and on about these aspects.
The emphasis is simply that there are few distinctive outright touchy-feely advances made in GPT-4 that can be readily seen and felt.
For those of you that favor GPT-4, and you certainly have good reasons to do so, I don’t want you to be leaping out of your chairs and exhorting that the aforementioned facets do a disservice to the advances made inside of GPT-4. I am merely pointing out that the real GPT-4 is a far cry from the hyped GPT-4, and that even the non-hyped GPT-4 is seemingly devised with few “sellable” distinctive features that would make it easier for the public at large to grasp why GPT-4 is so much better than ChatGPT.
Sometimes the sizzle is what sells the car.
An automotive engineer that works their heart and soul to improve an engine and drivetrain, allowing a car to speed up and shave a few split seconds on the 0 to 60 mph metric is likely a bit crestfallen when the public instead focuses on the shapely hood or the added roof rack. Sorry to say that the ornamented hood or the snazzy roof rack might end up being the keystone of why the car becomes a best seller, all else being equal.
That is human nature.
Beloved Feelings Toward ChatGPT Are Hard To Reinvent
Now that I’ve briefly covered the capabilities of GPT-4, let’s consider other factors that seem to underpin why GPT-4 is not as much a blockbuster in the eyes of the public as has been ChatGPT.
ChatGPT is the beloved AI that was the underdog and arose out of the blue.
The media hype about ChatGPT helped the OpenAI app become the darling of the world. People flocked to use it. ChatGPT became an overnight success. ChatGPT was heralded by the media and the public at large as an amazing breakthrough in AI.
The reality is that many other akin generative AI apps have been devised, often in research labs or think tanks, and in some cases were gingerly made available to the public. The outcome was not usually pretty. People prodded and poked at the generative AI and managed to get essays of an atrocious nature, see my coverage at the link here. The AI makers in those cases were usually forced to withdraw the AI from the open marketplace and revert to focusing on lab use or carefully chosen AI beta testers and developers.
Much of the rest of the AI industry was taken completely by surprise when ChatGPT managed to walk the tightrope of still producing foul outputs and yet not to the degree that public sentiment forced OpenAI to remove the AI app from overall access.
This was the true shock of ChatGPT.
Most people assumed the shock was the conversant capability. Not for those in AI. The surprise that floored nearly all AI insiders was that you could release generative AI that might spew out hateful speech and the backlash wasn’t fierce enough to force a quick retreat. Who knew? Indeed, prior to the release of ChatGPT, the rumor mill was predicting that within a few days or weeks at the most, OpenAI would regret making the AI app readily available to all comers. They would have to restrict access or possibly walk it home and take a breather.
The incredible success of the ChatGPT rollout has cautiously opened the door to other generative AI apps to also meet the street. For example, I’ve discussed the Google unveiling of Bard and how the Internet search engine wars are heating up due to a desire to plug generative AI into conventional web searching, see the link here.
Public sentiment about ChatGPT is incredibly strong, loyal, and so far long-lasting (I’ve discussed ways that this might eventually erode, see the link here).
We all especially seem to love those rags-to-riches stories. ChatGPT came out of nowhere. It was easy to use. It was fun to use. It amazed people with what it could do. It was made available for free. You could just go to a website, enter an email address, and voila, you were able to start using it. No hefty sign-up gauntlet. No slew of parameters to be set up. You simply log in, enter a prompt, and you get impressive essays as outputs.
The positivity that has gone from 0 to 60 in just a few months as associated with ChatGPT is a marketer’s dream. Take any other product such as say the iPhone by Apple, Coke from Coca-Cola, and other generally prized products and compare their perceived positivity to ChatGPT.
Okay, so does GPT-4 have or will it attain a comparable semblance of being beloved as is ChatGPT?
Maybe, but probably not.
GPT-4 is not an underdog. It is not going to have the same widespread cultural impact that ChatGPT has had. This is not a fault of GPT-4 per se, and merely the circumstances of the luck of timing and moment-in-time phenomena that ChatGPT has encountered.
Without the same divine touching of our hearts, GPT-4 is going to just need to slog it out like any other product in the AI marketplace. That’s the real world.
Tech-Awkward Naming Of GPT-4 Is A Big Catchiness Problem
One of the luckiest or possibly shrewdest aspects of ChatGPT was the naming of the AI app.
Prior to ChatGPT, the prevailing names of the OpenAI line of text-based generative AI products had a predominantly techie phrasing to them, for example, GPT-2, GPT-3, and GPT-3.5 are the classic versioning type of names that technologists love to use (side note, another one, InstructGPT, veered more closely to something friendlier in naming).
If ChatGPT had gone to market with the name of say GPT-3.5 or maybe a step-up as GPT-3.6 (made this up), this type of naming could have somewhat undercut the resulting popularity of the AI app. Simply put, the wording of “Chat” when combined with “GPT” turned out to be a stroke of good fortune. It is easy to pronounce, readily memorable, catchy, and maybe even cuddly.
Most people don’t know what GPT is an acronym for (generative pre-trained transformer), and they don’t especially care anyway. They viscerally know it probably is something scientific or technical in whatever it does mean. The word “Chat” is well-known, and we today often associate it with an entire category of online products known as chatbots. By combining the two together into the coined ChatGPT, you end up with a clever combination. It has the aura of technology due to the GPT, and it has simplicity and transparency due to the word Chat. Voila, the name has likely indelibly entered into our modern-day lexicon.
What about the name of its successor, GPT-4?
Some had assumed that OpenAI would almost certainly concoct a new name for GPT-4, doing so just before the launch date. Probably all manner of expensive specialty product-naming companies would have been consulted. Secret surveys and test groups would be assembled. Nobody would know the final chosen name until the big reveal at launch.
And that name is GPT-4.
The key point is that rather than having a lovable huggable name, we are back to the usual techie predisposition of using the shall we say internal insider name. This is not endearing to the public at large. Without a catchy name, gaining traction in the minds and hearts of the public can be challenging.
Be aware that there are counterarguments on this point.
For example, suppose that OpenAI came up with some super-duper name. This could have added to the splash and splendor of the GPT-4 launch. On the other hand, it could also have the downside of creating potential confusion in the marketplace and possibly undercutting the ChatGPT name. Having two cutesy names might be overloading the public consciousness. Maybe it makes more sense to keep the ChatGPT as the spotlighted star, and then slip the other one, the more drab GPT-4 name, into the world and see how things go.
Some are trying to rectify this naming conundrum by making up names for GPT-4. For example, you might see from time to time the made-up reference to “ChatGPT-4” whereby agitators have decided to create a new name that seems more likable and associated with the predecessor brethren of GPT-4. Many other variations exist.
The official name is nonetheless GPT-4.
Not especially lovable.
Just a straight-ahead techie-sounding name.
Complications Of Ready Availability For GPT-4 Is An Added Complication
Another bit of blockage to stardom consists of the availability complications regarding GPT-4.
Recall that ChatGPT was available for free at the initial launch, and you could sign-up with great ease (they did put a momentary cap on signups due to the high volume, but that probably added to the allure and was not a dampening or dent in the popularity).
Do you know how to get access to GPT-4?
Let me share the complicated path with you.
Perhaps the CEO of OpenAI, Sam Altman, summarized it best in his tweet on March 14, 2023, about the GPT-4 launch:
- “Here is GPT-4, our most capable and aligned model yet. It is available today in our API (with a waitlist) and in ChatGPT+.”
As indicated, there are currently two main ways to get access to GPT-4.
One means consists of signing up for ChatGPT+ (also known as ChatGPT Plus). Here’s the deal. Conventional access to ChatGPT is still free if you do the ordinary access route. If you want some additional ChatGPT usage perks, you can subscribe at $20 per month and become a ChatGPT Plus member. As a ChatGPT Plus member, you also variously have access to GPT-4 (depending upon how busy GPT-4 is).
Bottom-line is that to use GPT-4, you pretty much have to be willing to shell out twenty bucks per month. This shows you that the AI industry is learning vital lessons from Netflix, Disney+, and the streaming world.
The other primary method of getting access to GPT-4 consists of connecting to GPT-4 programmatically via the available API (application programming interface). Generally, here’s how that works. Someone else that has a piece of software that wants to leverage GPT-4 can have their software connect to GPT-4 via the API. You might for example have a sales-oriented package that could fruitfully use the capabilities of GPT-4. As such, you modify your sales package to include an API to GPT-4. I’ve covered aspects of the ample business opportunities in leveraging the API of generative AI such as ChatGPT in the link here.
I’ll be saying more about the GPT-4 API in a moment below. Hang in there.
The crux of this is that trying to use GPT-4 is a pain in comparison to the ease of using ChatGPT. If you want to try out GPT-4, you have to jump through hoops. Basically, it isn’t readily available for free. This obviously can dampen the volume of anticipated usage for now.
There are counterarguments associated with this consideration.
For example, you might say it is quite clever to use GPT-4 as a lure to get people to sign-up for the $20 per month ChatGPT Plus. People that might have been thinking about getting ChatGPT Plus are now provided with a powerfully attractive added bonus for doing so. Those sitting on the fence are probably convinced to subscribe at this juncture.
Another seemingly canny element consists of preventing the entire world from all at once trying to use GPT-4. If GPT-4 had been completely free and easy to use, the chances are that tons and tons of people would have signed up. The resulting avalanche of overloading would almost certainly have meant that many people weren’t going to have an uninterrupted experience. Whiners would have gotten front-page headlines. Even if GPT-4 was a miracle AI system, the carping over it being slow, sluggish, and impossible to access would have drowned out any other accolades.
One might argue compellingly that providing some barriers to using GPT-4 was prudent and crafty. The formula is apparent. Take a small ding for not having unfettered access but avert a mitigated reputational destroying disaster if everyone did have access.
Cumbersome Compliment That Microsoft Bing Uses GPT-4
Shortly after GPT-4 was unveiled, word soon spread that the OpenAI AI product that had been earlier melded into the Microsoft Bing search engine activity was GPT-4. For my earlier coverage of the Bing search engine and the generative AI addition, see the link here.
Here’s the official Bing announcement made online on March 14, 2023:
- “Congratulations to our partners at Open AI for their release of GPT-4 today. We are happy to confirm that the new Bing is running on GPT-4, which we’ve customized for search. If you’ve used the new Bing preview at any time in the last five weeks, you’ve already experienced an early version of this powerful model. As OpenAI makes updates to GPT-4 and beyond, Bing benefits from those improvements. Along with our own updates based on community feedback, you can be assured that you have the most comprehensive copilot features available.”
Is that pronouncement about having included GPT-4 something to be proud of and ought to be touted from the highest hilltops?
Probably not, depending upon one’s viewpoint (I’ll explain next).
You might have seen or heard in the news about all manner of crazy outputs that some have gotten the Bing search engine to devise when using the generative AI-added component. I’ve discussed how some people try to push the envelope and purposely get hate speech or other untoward outputs from generative AI, see the link here. There are those that do this to forewarn that generative AI is not ready for prime-time use. Others do it as a pastime or to garner attention. Etc.
The gist here is that Bing has already generally not had the utmost of reputations in terms of it being a search engine that has only a small portion of the search engine marketplace in comparison to Google, plus, the zany outputs that some got via the generative AI addition were also a bit of an embarrassment and letdown.
To associate the newly released GPT-4 with something that hasn’t gotten the most stellar of press is probably not the optimum advantage for the AI maker, even though it perhaps aids the search engine maker. Well, they are so closely allied now that it is probably a hair-splitting difference.
But I assume you get the essence of the predicament whereby GPT-4 garnered a somewhat cumbersome compliment.
Higher Cost Of API Use For GPT-4 Is A Challenge
I would guess that most people have no clue as to what the cost is to use the API for ChatGPT and nor what the cost to use the API for GPT-4 is.
There probably isn’t any notable reason they should need to know.
Only those that are desirous of using the APIs would likely figure this out. It is vitally important for them since they will need to somehow recoup the cost of using the APIs. If a sales package is going to use the API to access ChatGPT, this adds a new cost to the software maker of the sales package. They need to potentially increase their price to the users of the sales package to cover the cost or eat the cost in hopes of garnering added business for their sales package.
The reason I am going to bring it up here is that this is yet another small piece in the puzzle of what might keep GPT-4 from a lightning-neck pace of adoption. Money makes the world go round, and the same occurs if you want to use the API to connect to either ChatGPT or GPT-4.
Let’s first look at the API pricing for ChatGPT.
Per the official pricing on the OpenAI website, the cost to use the API of ChatGPT (known as gpt-3.5-turbo) is stated as $0.002 per 1,000 tokens.
That undoubtedly seems like a gobbledygook. A quick unpacking might help. When generative AI examines words, the words are divided into a set of tokens. Each token is roughly about 3 letters in size. The word “rabbit” which is six letters would generally be divided into two tokens “rab” and “bit”. In the English language, the average word length is about 4.7 letters. A handy rule of thumb is that you can multiply the number of tokens by approximately 75% to get the likely number of words. Thus, in the case of 1,000 tokens, the approximate equivalent is about 750 words.
Okay, with that under our belt, we will say that using the API of ChatGPT costs about $0.002 per 750 words. You might also find of interest that the average paragraph in the English language has around 200 words. So, for ease of discussion, let’s say that for $0.002 you can process about 4 paragraphs of normal size (750 / 200 = 3.75).
Things start to get a bit more complex as to how many paragraphs you might have when undertaking a written interactive dialogue. It all depends on what you are conversing about and how long of a conversation you want to have.
Imagine this. You start a conversation by entering a prompt that is a paragraph in size. ChatGPT emits an essay in response to your prompt. The essay is perhaps 5 paragraphs in size. You enter a new prompt that is one paragraph in size. ChatGPT responds with 6 paragraphs. You write two paragraphs as your next prompt. ChatGPT replies with 5 paragraphs. You close out the conversation.
At this juncture, you had consumed a total of 20 paragraphs, four at your prompts and sixteen as outputted by ChatGPT.
What did that cost?
If we assume that the price of $0.002 applies to about 4 paragraphs (roughly), and we used 20 paragraphs in this example, we know that the cost would approximately be (20 / 4) x $0.002 which is $0.01.
It cost a penny to use ChatGPT via the API for this rather short interactive conversation.
That seems like a heck of a deal. You can envision why many are flocking to add the use of ChatGPT to their software packages. Of course, you need to be mindful of this example. Suppose you have a user that is accessing the API via your software package and they carry on a full-blown written interactive conversation. Let’s suppose that might be a dime in cost.
If you have a thousand users of your package and they all are hourly using the API to the same degree as the dime, you are incurring perhaps a $100 added cost per hour for the ChatGPT usage. My point simply is that the numbers can add up due to volume and frequency. Be cautious to not get yourself into a bind.
I trust that this has whetted your appetite to know what the cost to use GPT-4 via the API is.
First, remember that I had just moments earlier stated that the official pricing to use the API of ChatGPT (known as gpt-3.5-turbo) is stated as $0.002 per 1,000 tokens.
The price to use GPT-4 is either $0.06 per 1,000 tokens for the 32K context, or it is $0.03 per 1,000 tokens for the 8K context (the context refers to the overall size of the contextual discussion taking place when interacting with GPT-4, wherein the larger the context the greater the cost).
An easier way to see these prices is side-by-side:
- $0.002 per 1,000 tokens (gpt-3.5-turbo)
- $0.060 per 1,000 tokens (GPT-4 at 32K context)
- $0.030 per 1,000 tokens (GPT-4 at 8K context)
This is quite a pricing leap.
Consider that the $0.002 pricing goes up by 30x to use the topmost GPT-4 (32K context) and goes up about 15x to use the other GPT-4 (8K context).
Another angle would be to say that my earlier example of a penny for using the API of ChatGPT would be thirty cents for the topmost GPT-4 (32K context) or around fifteen cents for the other GPT-4 (8K context). A thousand users on an hourly basis of added cost might jump from $100 per hour to $3,000 per hour or $1,500 per hour, though again that’s just one example and you need to consider the volume and frequency for whatever usage you have in mind.
I’ve noted in my column coverage on these matters that the decision to use the API of ChatGPT versus the API of GPT-4 is not a slam-dunk. You need to closely estimate the nature of the usage that might occur. You also need to weigh the advantages of GPT-4 over ChatGPT and whether the added cost is worth the benefits you expect to incur.
All in all, I’ve already predicted that we are going to see a flood of software that will use either ChatGPT or GPT-4 via the API. Some will probably start with ChatGPT to see how the marketplace reacts to the added connection in their package. At some later point, they are likely to upgrade to GPT-4. Others will potentially opt to skip past ChatGPT and decide that the cost is low enough overall to warrant the immediate aim of using GPT-4 and not play around with ChatGPT. If the potential profit or added market share to your package warrants the higher API cost, so be it.
Keep in mind that your mileage may vary and you need to dutifully figure out which path, if either one, makes sense for your software package. You ought to also look at other generative AI apps and review their pricing and features too.
Being Able To Pass Examines Is Not Especially A Public Windfall
We are getting toward the end of this list of points to consider, so I’ll speed things up.
You might have seen or heard that GPT-4 was able to pass various exams that are considered college-going or college-graduating types of tests. For example, GPT-4 was tried out on the SAT exam, the GRE exam (used for graduate college admissions), numerous high school AP exams that can earn college credits, and other exams. The most noted one of them all was probably the feared and esteemed bar exam that is customarily used to test lawyers when seeking to practice law (known as the Uniform Bar Exam). For my in-depth discussion on the bar exam passing and how this impacts lawyers and legal services, see the link here and the link here, just to name a few.
To those inside the AI industry, this was an impressive feat.
To the everyday public, this is somewhat impressive but maybe not quite as much as one might assume. People generally tend to think of “test-taking” as a somewhat narrow skill. Everybody knows that written tests are not the grand end-all. Sure, it is notable and possibly dramatic that a generative AI was able to do so well on those formidable exams. Nobody reasonably disputes that achievement.
The thing is, how does that translate into doing real-world tasks of a daily nature?
I would suggest that passing written exams is probably not going to win the hearts and minds of the public to the use of generative AI. It is a convenient metric for those in AI. It is measurable. You can report the results without much ambiguity. And so on.
For the public, likely “proof” of generative AI capabilities will require other or at least additional forms of triumphant accomplishments.
Multi-Modal Requires Delayed Gratification Right Now
I’ve predicted that this year we are going to see the advent of multi-modal generative AI, and it will astound and knock the socks off of everyone (for more on this, visit the link here).
Right now, we have essentially singular-mode generative AI.
For example, ChatGPT takes text as input and produces text as its output.
Think of it this way:
- ChatGPT Input: Text
- ChatGPT Output: Text
Some generative AI apps take text as input and produce images or artwork as output. In that case, the input is just one mode, namely text, while the output is also just one mode, namely an image or artwork.
Think of it this way:
- Generative AI Input: Text
- Generative AI Output: Image or artwork
With me on this so far?
I hope so.
AI makers are burning the midnight oil to try and make their generative AI multi-modal.
The concept is simple. For input, you might have text, plus you might also have the entry of an image or artwork. That proffers two modes of input. The output might consist of generated text, plus a generated image or artwork, and let’s suppose a generated video too. That would be three modes of output.
The grandiose version of all feasible modes would be this:
- Generative AI Input: Text
- Generative AI Input: Image or artwork
- Generative AI Input: Audio
- Generative AI Input: Video
- Generative AI Input: Other modes
- Generative AI Output: Text
- Generative AI Output: Image or artwork
- Generative AI Output: Audio
- Generative AI Output: Video
- Generative AI Output: Other modes
That would be the pinnacle of multi-modal generative AI. You would have all modes available for input as prompts, and you would likewise have all modes available for the generated output. The user would be able to make their preferred choices, doing so at their whim.
This is where we are headed.
It is going to be breathtaking.
From an AI Ethics perspective, this is also going to be worrisome. If you thought deepfakes were a problem now, wait until we have full multi-modal generative AI. Hold onto your hats. You might also anticipate that lawmakers and regulators will inexorably be drawn into the generative AI marketplace when all sorts of unsavory sneakiness happen by evildoers exploiting the multi-model capabilities. New AI Laws will gain special urgency.
Back to the matter at hand, GPT-4 was launched and lauded as being multi-modal.
Here is what the official OpenAI GPT-4 Technical Report indicates:
- “We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs.”
In summary, you can enter text and you will get outputted text, plus you can possibly enter an image at the input.
This is what GPT-4 provides:
- GPT-4 Input: Text
- GPT-4 Input: Image or artwork (this functionality is not yet released to the public)
- GPT-4 Output: Text
Compare this to the grand aims of true multi-modal generative AI with all manner of inputs and all manner of outputs (i.e., GPT-4 is a rather abbreviated list at this time, as I’ll mention further in a moment).
Demonstrations showcasing the image or vision processing of inputted images have indicated that the items in a picture for example could be identified by the generative AI and then composed into a written narrative explaining the picture. You can ask the generative AI to explain what the picture seems to depict. All in all, the vision processing will be a notable addition.
The vision processing or image analysis capability is not yet available for public use (per the OpenAI website blog):
- “To prepare the image input capability for wider availability, we’re collaborating closely with a single partner to start.”
The crux of all of this is that it is heartwarming to realize that GPT-4 apparently does have the capability to do image input and analysis. Many are eagerly awaiting the public release of this feature. Kudos to OpenAI for nudging into the multi-modal arena.
So, we have text as input, plus image as input (when made available for public use), and text as output.
Some though have been handwringing in the AI community that this barely abides by the notion of multi-modal. Yes, there is one more mode, the image as input. But not an image as output. There seemingly isn’t audio as input, nor audio as output. There seemingly isn’t video as input, nor video as output. Those with a smarmy bent find this to be “multi-modal” in the most minimalist of ways.
The counterargument is that you have to crawl before you walk, and walk before you run.
We can ponder this seeming “multi-modal” capability from a skeptical or cynical perspective. Remember that I earlier said that when a carmaker comes out with a new car they want to have some shiny new features or functionality that can be marketed to the hilt. I also suggested that GPT-4 didn’t especially have any standout new features or functionality, at least none that would set it apart in a very distinctive way from ChatGPT in the eyes of the general public.
Aha, the provision of images as inputs is in fact the kind of added feature that would potentially stand out. A cynic might say that the AI maker nearly had to announce the image processing capability, regardless of what its status might be. This was the one new thing that they could hang their hat on and that would be a clear-cut standout from before. A cynic might further contend that showing demos is the classic demoware form of vendor mischief and so too is the deferring tactic of indicating that the feature is being polished before being released.
How much street cred or credit should be given to a product feature that is demonstrated but not released into the hands of the general public?
Some have insisted that this is perfectly acceptable, done all the time, and you have to simply hold your horses and wait for that delayed gratification to ensue.
One added twist is to contemplate the intriguing matter of what would have happened had the image processing been released at the time of launch. In a sense, that’s a difficult choice. It could propel the launched AI app to a greater feverish pitch as people scrambled to try this new nifty feature. The downside though is that since this is a new element and presumably untested by the public all told, the kind of zany things that people might have done could have become a public relations nightmare.
The thorny decision might have been that darned almost for sure if you do, but only slightly darned if you don’t.
Two gladiators enter the ring and both successfully walk out, shaking friendly hands. They are each and collectively entirely happy to be able to coexist.
That is a more fitting trope for the status of ChatGPT and the recent launch of GPT-4.
Some had wondered whether GPT-4 would instantly cannibalize the market for ChatGPT. If GPT-4 was leaps and bounds beyond ChatGPT, the belief was that ChatGPT would be left in the dust. On the other hand, and though not at all expected, if GPT-4 was somehow inferior to ChatGPT, the dust might be where GPT-4 would end up landing.
Nope, they each have their own particular tradeoffs, consisting of features and functions, along with differences in costs and availability.
If ChatGPT could talk, maybe it would be saying hello and warmly welcoming its brethren GPT-4. Likewise, GPT-4 would be complimenting ChatGPT on a job well done and appreciative of having set the stage for GPT-4 to enter into the real world from the labs that it has long been under construction.
A final remark for now.
Mother Teresa famously said this: “I can do things you cannot, you can do things I cannot; together we can do great things.”
At this time, it seems like the same could be said of the relationship between ChatGPT and GPT-4. They are respectful family members that each have their own path and particular destiny.