Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT-4

openai.com

Post reply on HN
Print view View original

581–590 of 1001 posts

Re: GPT-4

#581

Post by barking_biscuit »

hospitalJail wrote: post #531
aabajian wrote: post #456

I'll be finishing my interventional radiology fellowship this year. I remember in 2016 when Geoffrey Hinton said, "We should stop training radiologists now," the radiology community was aghast and in-denial. My undergrad and masters were in computer science, and I felt, "yes, that's about right." If you were starting a diagnostic radiology residency, including intern year and fellowship, you'd just be finishing now.…

If you are in the US. It is more important to have the legal paperwork, than to be factually correct. The medical cartels always will get their cut.

username checks out

Re: GPT-4

#582

Post by elicksaur »

BoiledCabbage wrote: post #255
Imnimo wrote: post #218

A class of problem that GPT-4 appears to still really struggle with is variants of common puzzles. For example: >Suppose I have a cabbage, a goat and a lion, and I need to get them across a river. I have a boat that can only carry myself and a single other item. I am not allowed to leave the cabbage and lion alone together, and I am not allowed to leave the lion and goat alone together. How can I safely get all three…

It's a good observation. Although on the flip side, I almost went to type up a reply to you explaining why you were wrong and why bringing the goat first is the right solution. Until I realized I misread what your test was when I skimmed your comment. Likely the same type of mistake GPT-4 made when "seeing" it. Intuitively, I think the answer is that we do have two types of thinking. The pattern matching fast thinkin…

> Until I realized I misread what your test was when I skimmed your comment. Likely the same type of mistake GPT-4 made when "seeing" it.

Wouldn’t we expect a computer program with perfect knowledge of the input to be less likely to make such a mistake? You made that mistake because you didn’t actually read the whole prompt, but I would expect GPT to take into account every word.

Really it shows that it doesn’t actually have a model of these objects. It can mimic knowing what a lion is, but it doesn’t actually have the concept of a lion or cabbage being an actual singular item, so its program mistracks what is an item and what the rules about an item are in the given prompt.

Re: GPT-4

#583

Post by sebzim4500 »

pwpw wrote: post #267
nealabq wrote: post #227

Test taking will change. In the future I could see the student engaging in a conversation with an AI and the AI producing an evaluation. This conversation may be focused on a single subject, or more likely range over many fields and ideas. And may stretch out over months. Eventually teaching and scoring could also be integrated as the AI becomes a life-long tutor. Even in a future where human testing/learning is no l…

I think a shift towards Oxford’s tutorial method [0] would be great overall and compliments your point. “Oxford's core teaching is based around conversations, normally between two or three students and their tutor, who is an expert on that topic. We call these tutorials, and it's your chance to talk in-depth about your subject and to receive individual feedback on your work.” [0] https://www.ox.ac.uk/admissions/under…

We had something similar in Cambridge and it was extremely useful. I can't imagine how the course would have worked without it, honestly.

If AI can achieve this (and honestly I do not think GPT-4 is far off, at least for primary and middle school level stuff) it will be a far bigger win for education than the internet was.

Re: GPT-4

#584

Post by hooande »

After watching the demos I'm convinced that the new context length will have the biggest impact. The ability to dump 32k tokens into a prompt (25,000 words) seems like it will drastically expand the reasoning capability and number of use cases. A doctor can put an entire patient's medical history in the prompt, a lawyer an entire case history, etc.

As a professional...why not do this? There's a non-zero chance that it'll find something fairly basic that you missed and the cost is several cents. Even if it just phrases something obvious in a way that makes you think, it's well worth the effort for a multimillion dollar client.

If they further increase the context window, this thing becomes a Second Opinion machine. For pretty much any high level job. If you can put in ALL of the information relevant to a problem and it can algorithmically do reasoning, it's essentially a consultant that works for pennies per hour. And some tasks that professionals do could be replaced altogether. Out of all the use cases for LLMs that I've seen so far, this seems to me to have the biggest potential impact on daily life.

edit (addition): What % of people can hold 25,000 words worth of information in their heads, while effectively reasoning with and manipulating it? I'm guessing maybe 10% at most, probably fewer. And they're probably the best in their fields. Now a computer has that ability. And anyone that has $20 for the OpenAI api can access it. This could get wild.

Re: GPT-4

#585

Post by WoodenChair »

Imnimo wrote: post #218

A class of problem that GPT-4 appears to still really struggle with is variants of common puzzles. For example: >Suppose I have a cabbage, a goat and a lion, and I need to get them across a river. I have a boat that can only carry myself and a single other item. I am not allowed to leave the cabbage and lion alone together, and I am not allowed to leave the lion and goat alone together. How can I safely get all three…

> Even after I pointed this mistake out, it repeated exactly the same proposed plan. It's not clear to me if the lesson here is that GPT's reasoning capabilities are being masked by an incorrect prior (having memorized the standard version of this puzzle) or if the lesson is that GPT'S reasoning capabilities are always a bit of smoke and mirrors that passes off memorization for logic.

It has no reasoning capabilities. It has token prediction capabilities that often mimic reasoning capabilities.

Re: GPT-4

#586

Post by bpicolo »

arrosenberg wrote: post #112
_boffin_ wrote: post #70

This technology has been a true blessing to me. I have always wished to have a personal PhD in a particular subject whom I could ask endless questions until I grasped the topic. Thanks to recent advancements, I feel like I have my very own personal PhDs in multiple subjects, whom I can bombard with questions all day long. Although I acknowledge that the technology may occasionally produce inaccurate information, the…

I'm very excited for the future wave of confidently incorrect people powered by ChatGPT.

You're going to get confidently incorrect arguments on the internet straight from ChatGPT without the human filter.

Re: GPT-4

#587

Post by redbell »

ml_basics wrote: post #136

From the paper: > Given both the competitive landscape and the safety implications of large-scale models like GPT-4, this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar. I'm curious whether they have continued to scale up model size/compute significantly or if they have managed to make significant innovati…

> this report contains no further details about the architecture (including model size), hardware, training compute

As a beginner in the NLP world, this may serve me a purpose which is to hide the complexity behind building such models.. numbers like xyzB parameters, 12K A100s.. are scary, so I still can dream of building one system one day. This story [0] and this one [1] hide some extremely complex edge cases that a beginner will never though of or had the courage to start if he knew what is the real cost.

We may, however, still be able to infer some details [probably in the future] knowing how Microsoft had re-arranged its infrastructure to welcome OpenAI training [2]

_________________

[0]. https://www.construct.net/en/blogs/ashleys-blog-2/simple-sof...

[1]. https://prog21.dadgum.com/29.html

[2]. https://www.theverge.com/2023/3/13/23637675/microsoft-chatgp...

Re: GPT-4

#588

Post by barking_biscuit »

chis wrote: post #46
teruakohatu wrote: post #26

Access is invite only for the API, and rate limited for paid GPT+. > gpt-4 has a context length of 8,192 tokens. We are also providing limited access to our 32,768–context (about 50 pages of text) version, gpt-4-32k, which will also be updated automatically over time (current version gpt-4-32k-0314, also supported until June 14). Pricing is $0.06 per 1K prompt tokens and $0.12 per 1k completion tokens. The context le…

I'm really curious to see if expanding the context length this much will allow GPT to do typical software development tasks on a big codebase. If it can take in a github issue and produce decent code solving a complex issue across many files... will certainly be an interesting time.

>If it can take in a github issue and produce decent code solving a complex issue across many files... will certainly be an interesting time.

Oh snap. I didn't even think about that!

That gives me a fun idea!

I've got a repo that I built and setup CI/CD and setup renovate to automatically upgrade dependencies and merge them when all the tests pass, but of course sometimes there are breaking changes. I don't actively work on this thing and hence it's just got issues sitting there when upgrades fail. It's the perfect testing ground to see if I can leverage it to submit PRs to perform the fixes required for the upgrade to succeed! That'll be hectic if it works.

Re: GPT-4

#590

Post by swatcoder »

mk_stjames wrote: post #577

I just finished reading the 'paper' and I'm astonished that they aren't even publishing the # of parameters or even a vague outline of the architecture changes. It feels like such a slap in the face to all the academic AI researchers that their work is built off over the years, to just say 'yeah we're not telling you how any of this is possible because reasons'. Not even the damned parameter count. Christ.

In the old days of flashy tech conferences, that was precisely the sign of business-driven demo wizardry.

The prerecorded videos, the staff-presented demos, the empty hardware chassis, the suggestive technical details, etc

They have “reasons” for not giving away details, but there are good odds that the ultimate reason is that this is a superficial product update with a lot of flashy patchwork rather than that fundamental advance in AI technology we’d assume from the name.

Post reply on HN