OpenAI Releases 'Smarter, Faster' ChatGPT - Plus $200-a-Month Subscriptions for 'Even-Smarter Mode' (venturebeat.com) 56
Wednesday OpenAI CEO Sam Altman announced "12 Days of OpenAI," promising that "Each weekday, we will have a livestream with a launch or demo..." And sure enough, today he announced the launch of two things:
- "o1, the smartest model in the world. Smarter, faster, and more features (e.g. multimodality) than o1-preview. Live in ChatGPT now, coming to API soon."
- "ChatGPT Pro. $200/month. Unlimited usage and even-smarter mode for using o1. More benefits to come!"
Altman added this update later: For extra clarity: o1 is available in our plus tier, for $20/month. With the new pro tier ($200/month), it can think even harder for the hardest problems. Most users will be very happy with o1 in the plus tier!
VentureBeat points out that subscribers "also gain access to GPT-4o, known for its advanced natural language generation capabilities, and the Advanced Voice feature for speech-based interactions."
And even for non-subscribers, ChatGPT can now also analyze images, points out VentureBeat, "a hugely helpful feature upgrade as it enables users to upload photos and have the AI chatbot respond to them, giving them detailed plans on how to build a birdhouse entirely from a single candid photo of one, for one fun example." In another, potentially more serious and impressive example, it is now capable of helping design data centers from sketches... o1 represents a significant evolution in reasoning model capabilities, including better handling of complex tasks, image-based reasoning, and enhanced accuracy. Enterprise and Education users will gain access to the model next week... OpenAI's updates also include safety enhancements, with the o1-preview scoring 84 on a rigorous safety test, compared to 22 for its predecessor...
To encourage the use of AI in societal-benefit fields, OpenAI has announced the ChatGPT Pro Grant Program. The initiative will initially award 10 grants to leading medical researchers, providing free access to ChatGPT Pro tools.
In a video Altman displays graphs showing o1 dramatically outperforms gpt4o on math questions, on competition coding at CodeForces, and on PhD-level science questions.
- "o1, the smartest model in the world. Smarter, faster, and more features (e.g. multimodality) than o1-preview. Live in ChatGPT now, coming to API soon."
- "ChatGPT Pro. $200/month. Unlimited usage and even-smarter mode for using o1. More benefits to come!"
Altman added this update later: For extra clarity: o1 is available in our plus tier, for $20/month. With the new pro tier ($200/month), it can think even harder for the hardest problems. Most users will be very happy with o1 in the plus tier!
VentureBeat points out that subscribers "also gain access to GPT-4o, known for its advanced natural language generation capabilities, and the Advanced Voice feature for speech-based interactions."
And even for non-subscribers, ChatGPT can now also analyze images, points out VentureBeat, "a hugely helpful feature upgrade as it enables users to upload photos and have the AI chatbot respond to them, giving them detailed plans on how to build a birdhouse entirely from a single candid photo of one, for one fun example." In another, potentially more serious and impressive example, it is now capable of helping design data centers from sketches... o1 represents a significant evolution in reasoning model capabilities, including better handling of complex tasks, image-based reasoning, and enhanced accuracy. Enterprise and Education users will gain access to the model next week... OpenAI's updates also include safety enhancements, with the o1-preview scoring 84 on a rigorous safety test, compared to 22 for its predecessor...
To encourage the use of AI in societal-benefit fields, OpenAI has announced the ChatGPT Pro Grant Program. The initiative will initially award 10 grants to leading medical researchers, providing free access to ChatGPT Pro tools.
In a video Altman displays graphs showing o1 dramatically outperforms gpt4o on math questions, on competition coding at CodeForces, and on PhD-level science questions.
But will it help me (Score:5, Funny)
Get first Post?
Re:But will it help me (Score:5, Funny)
I not only got first post, I also saved $200 a month by not subscribing.
Re: (Score:2, Insightful)
So the answer to your question was "no".
Re: (Score:2)
First post is nothing special anymore, sadly.
Re: (Score:2)
Makes me wonder what his definition of "smart" is. I'm beginning to think, at the rate that each version increases hallucinations instead of decreasing them, that the definition of smart the creators were going for was "pulling stuff out of my ass to attract venture capital".
PhD level science questions? (Score:4, Insightful)
And no, by "research", I do not mean googling.
Literally the whole point of a scientific PhD is to perform experiments and study to answer a specific research question that no one has looked into yet.
Whilst ChatGPT can probably can answer "PhD-level science questions" with the same generation of plausible bullshit it answers all questions, I very much doubt ChatGPT can answer PhD-level science questions with any sort of accuracy.
It can't do that without performing experiments (that in some cases might be complex enough to last years).
Just more of the marketing BS silicon valley seems to be full of these days. Remember when California was actually making products that benefited society as well as making money?
Re: (Score:2, Interesting)
I have colleagues who have started using chatGPT to solve research-related mathematical questions. The comments I have heard is that it works and is more convenient than having to search again for the syntax of (Wolfram) Mathemetica.
Re:PhD level science questions? (Score:4, Interesting)
Re: (Score:2)
It isn't assisting with syntax, it fully replaces computation software such as Mathematica or Maple by resolving equations entirely. It certainly won't make the experiment for us, but it solves the theoretical part.
Re: (Score:3)
It isn't assisting with syntax, it fully replaces computation software such as Mathematica or Maple by resolving equations entirely. I
No. It cannot do that. It may be able to do it for often documented examples, but that is it. That said, some AI tools (e.g. ChatGPT) actually have a link to Wolfram Alpha, and if you trigger that, you get that answer from Wolfram Alpha and not an AI answer.
Re: (Score:3)
Given it's history with hallucinations, I wouldn't trust it to solve 2+2=4
Re: (Score:1)
Re:PhD level science questions? (Score:4, Interesting)
It was not a good idea to blindly follow instructions found on one single source; but it is still not a good idea to give up the a tool entirely. It has uses, you need to find where it fits in your workflow. It saves a lot of time in some scenarios when used adequately.
Re: (Score:3)
I disagree. It is perfectly fine to follow instructions in the datasheet of the manufacturer in this case. For arbitrary sources, no number of them combined is safe.
Re: (Score:2)
Re: PhD level science questions? (Score:2)
They wonâ(TM)t settle because it will open their exposure for every wrong answer. Right now they have no liability and they want to keep it that way.
Re: (Score:2)
Sure. But are these mathematical research questions? No, they are not. Incidentally, "Where is the coffee machine?" is very much a research-related question in that its answer enables research, but it is in no way a research question.
Re: (Score:2)
Well, I guess they've gotta please the investors & shareholders somehow.
Fake it 'till you make it or 'till the VC money runs out! (More than likely the
Re: (Score:3)
It can answer anything accurately if you give it the answers.
Re: (Score:3)
Indeed. A PhD level science questions requires the qualification to do that PhD to answer competently. Any claim that an AI tool can do it is, at this time, a complete, bald-faced lie. Incidentally, if it is a well-understood question, it is not PhD-level anymore.
Re: (Score:3)
I think the definition of Silly Con Valley Intelligence is "Making stuff up to attract venture capital".
By that definition of intelligence, ChatGPT super smart mode is likely running at a rate of 500 lies a minute.
Re: (Score:2)
No, but that's not the point. What it should be able to do is find out if your thesis has been done before and find and document all the references for you thus freeing you up to do the experiments. AI is part of the natural progression in information access. A physical library has severe limitations. You have to spend a lot of time looking through the card catalog hoping to find something relevant. Then you have to hope and pray that some asshole hasn't checked out the book you need when you need it. Yo
Re:What the hell are they smoking? (Score:5, Insightful)
I haven't even found Chat GPT to be useful yet. How do they expect me to pay $200/month for something that may be useful when they haven't proven its worth yet?
They don't expect anything of you at all. They expect some people who already use ChatGPT frequently - and who bump up against its limitations - to pay for a version which promises fewer of those limitations.
Those people who rely on ChatGPT to do worthwhile work on a regular basis DO exist. Some of them comment here pretty frequently.
Re: (Score:2)
There is, unfortunately for everyone involved, an option apart from "paying more money will make the AI more accurate"
Re: (Score:3)
Paying for information (and paying extra for more correct/more detailed data) is a long-established practice. See here [youtube.com] an ancient exemplification of the tradition that culminated in ChatGPT.
Re: (Score:2, Insightful)
You just proved my point. "Paying for more correct data" is the problem here. There's no such thing as "more correct data". Data is either correct in whole, or incorrect, even if in part. Chat GPT has rarely given me correct data. If it had, then I can see paying for more options or more detailed data, but until I know it can get things right, there's no point in paying for it. Perhaps the people who are getting correct data more often than me from it have asked trivial questions.
Re:What the hell are they smoking? (Score:5, Interesting)
The thing about GPT LLMs is that they're extremely susceptible to the "streetlight effect" (See: https://en.wikipedia.org/wiki/... [wikipedia.org]), among many other well-known threats to validity already identified by science.
The many ways in which GPT LLMs can be wrong are complex & subtle & require a very high level of expertise to identify & mitigate them & also to decide when they're inappropriate for a task. I reckon that's why, according to preliminary reports, experts can improve their productivity by using them but regular workers end up with poorer quality & lower productivity.
In other words, you need to be an expert in your field to be able to benefit from using GPT LLMs.
Re: (Score:2)
Data is neither correct nor incorrect. It's just data.
That's just semantics and beating around the bush. What you're saying is that I ask CG a question and it gives me "data". I need to know whether the data is something I can rely on (meaning that it's correct) or if I can't (meaning that it's incorrect). If I ask CG to solve a quadratic equation for me and it gives me only one of the two answers, then that "data" is incorrect (or rather incomplete). Saying if you give us $200/month, we'll give you more correct data doesn't instill in me the confidence I
Re: (Score:2)
There is no chance of the current approach to AI producing "more correct" information. More detailed, perhaps, but each of those details will still have a 40% chance of being a total hallucination.
Re: (Score:2)
And this is the crux of the problem for me and something I can't wrap my head around. If I know there's a 40% (or whatever) chance that what I'm getting back from the LLM is wrong in cases where I'm expecting a discrete answer, or hallucination in other cases, then that means I'd have to verify everything I get back from the LLM because I can't rely on its answers.
Re:What the hell are they smoking? (Score:5, Interesting)
I haven't even found Chat GPT to be useful yet. How do they expect me to pay $200/month for something that may be useful when they haven't proven its worth yet?
They don't expect anything of you at all. They expect some people who already use ChatGPT frequently - and who bump up against its limitations - to pay for a version which promises fewer of those limitations.
Those people who rely on ChatGPT to do worthwhile work on a regular basis DO exist. Some of them comment here pretty frequently.
Yup, I am one of them, I use it for some undergraduate math classes I teach. Earlier today I needed to give a student a final exam early. The issue is, that spills the final exam. Since my exams are all written in LaTeX (along with the solutions) I can just feed the exam into o1 and ask for a new version with an updated key, ensuring problems work out similarly. It caused me zero stress to need a new version of that exam, because I know the whole process of getting a new one will take under 5 minutes. Re-writing a 20 question final exam for math in LaTeX, along with a key, and ensuring the problems work out well, takes vastly longer than 5 minutes manually. Additionally, when I want student handouts, I feed the LaTeX code for my Beamer presentations into o1, and it'll solve all of the problems in the presentation, re-format into a handout, and provide the LaTeX code. Viola! Students have fully worked examples, and everything in my Beamer slides. It takes time correcting errors (which are common) but that's far easier than doing it from scratch. Consequently, the promise of fewer errors to correct is truly giving me some o1-pro envy.
Alas, I'm not in a profession that pays enough for my to shell out that kind of money, so I won't do it. However, I do have a ton of material to create for two new classes I'm teaching next semester, so I'm tempted to get a month subscription to churn out everything I need, and then cancel it. But $200 is pretty step. It's ashame they can't have something like a "1-day pass" for people that need to get a particular big job done, and don't really need it for the whole month.
Re: (Score:2)
It takes time correcting errors (which are common) but that's far easier than doing it from scratch.
Can you elaborate? E.g. a percentage of output tokens you need to change, types of errors it does more commonly, does it deteriorate with context size, at which levels, ...
Re: (Score:2)
Alas, I'm not in a profession that pays enough for my to shell out that kind of money, so I won't do it. However, I do have a ton of material to create for two new classes I'm teaching next semester, so I'm tempted to get a month subscription to churn out everything I need, and then cancel it. But $200 is pretty step. It's ashame they can't have something like a "1-day pass" for people that need to get a particular big job done, and don't really need it for the whole month.
You will still get everything you are used to with 20 dollars, even the new gpt-o1. That 200 is for having even more advanced model (or, I suspect, just having more iterations with the same one).
Are LLM's the new calculator? (Score:2)
I recall an undergraduate publishing design class I had where the professor required u
Re: (Score:2)
But $200 is pretty step. It's a shame they can't have something like a "1-day pass" for people that need to get a particular big job done, and don't really need it for the whole month.
You have many degrees and you can't understand how late stage capitalism works.
Re: (Score:2)
I haven't even found Chat GPT to be useful yet. How do they expect me to pay $200/month for something that may be useful when they haven't proven its worth yet?
They don't expect anything of you at all. They expect some people who already use ChatGPT frequently - and who bump up against its limitations - to pay for a version which promises fewer of those limitations.
Those people who rely on ChatGPT to do worthwhile work on a regular basis DO exist. Some of them comment here pretty frequently.
> Some of them comment here pretty frequently.
Maybe he doesn't trust comments generated by ChatGPT /s
Re: (Score:2)
I haven't even found Chat GPT to be useful yet. How do they expect me to pay $200/month for something that may be useful when they haven't proven its worth yet?
They don't expect anything of you at all. They expect some people who already use ChatGPT frequently - and who bump up against its limitations - to pay for a version which promises fewer of those limitations.
Those people who rely on ChatGPT to do worthwhile work on a regular basis DO exist. Some of them comment here pretty frequently.
> Some of them comment here pretty frequently.
Maybe he doesn't trust comments generated by ChatGPT /s
Thanks - I needed a laugh just now!
Re: (Score:3)
At $200 a month they'd better remove the "will not write violent feminist porn" limitation. Because that's the only class of people who will be using it.
Throw in some swag and, nope, still not gonna (Score:2)
At the very least it should come with something you can wear to capitalize on that whole conspicuous consumption thing. Perhaps a t-shirt which says "I spend $200/mo for ChatGPT and all I got was this stupid t-shirt".
Re: (Score:2)
I found one use so far: When you search for a thing but do not know the correct name, ChatGPT is usually actually helpful based on a description. Once you have the correct name, you can do a conventional search for that with better results.
Re: (Score:2)
You haven't found it useful, because you haven't been paying $200/month yet, obviously. /s
Seriously, I have found plenty of uses for ChatGPT, but certainly not $200 worth of uses. It speeds us searches a little bit. Rather than spending time looking for the right page on a search engine, you now spend your time exploring all the wrong answers GPT gives you, to rule them out. Turns out it's still faster to do it that way
Re: (Score:2)
I have hit the limit sometimes when doing a lot of design work, so I can see $200 being a ticket to not being told to wait 5-15 minutes, especially with the higher end LLMs. If I were doing my hobbies full time for income, I'd probably plunk down the $200 for unlimited AI goodness. Especially when designing OpenSCAD stuff.
The more you pay, the better the answers (Score:2)
Now that is something for a new business model.... =/
Re: (Score:2)
Re: (Score:2)
Well, that the answers are "better" is a marketing claim. Maybe the hallucinations are simply more credible...
I find copilot useful for UML generation (Score:2)
I find copilot (which I think runs on GPT4 Turbo) useful in generating UML diagrams (via plantuml), you can give it source code, or a natural language description of the diagram. I have got good results so far with this. It will be interesting to see if the latest version is any good, but 200 bucks a month means I wont be doing that anytime soon...
Re: (Score:2)
But are UML diagrams a benefit or a problem? I would submit that it very much depends.
Re: (Score:2)
So full cretin $200? (Score:2)
And "complete cretin" cheaper? Wow, where do I sign up?
The magic words are "Are you sure ?" (Score:3)
Ask ChatGPT about something sufficiently obscure, and it will just completely make up the answer, sometimes giving a very long list of things that don't apply, but will take you a precious amount of time to investigate. Asking the magic words is often enough for it to completely reverse course.
So, for $200/month, is it smart enough to ask itself that question ?
Does the increased "intelligence" of the LLM make (Score:2)
ChatGPT speaketh with forked tongue (Score:2)
b. ChatGPT spouts too much Wokism.
Is this April 1st? (Score:2)