OpenAI Rolls Out GPT-4.5

OpenAI released an early version of its new AI model GPT-4.5 to select users on Thursday, following development challenges that delayed the project last year. The Microsoft-backed startup said the new model responds better to subtle cues in written prompts and excels at chatting, writing and coding. OpenAI expects it will produce fewer fabricated responses than previous versions.

Initially available as a "research preview," access is limited to software developers and users who pay $200 monthly for ChatGPT Pro subscriptions. The company plans to gather feedback before wider distribution. According to OpenAI's blog post, GPT-4.5 will be the company's last model that doesn't use additional computing power to analyze queries before responding. Future releases will incorporate the reasoning approach already used in its newer models like o1 and o3.

Anti-snake is the least subtile of the beasts.

[Pseudonymous Powers]

"The Microsoft-backed startup said the new model responds better to subtle cues..."

It wasn't subtle! I told you to crawl up your own ass and die there!

And the constant delivery scam continues

[gweihir]

Oh, and look, it is even expected (!) to deliver less lies and made-up crap. Nice.

Re:

[KingStannis]

Fewer.

Re:

[nightflameauto]

I think he meant "lesser." As in "he ate the fudge cookie, not the chocolate chip," is a lesser lie than, "she killed the whole family, not just the fish in the living room aquarium."

Re:

[gweihir]

Actually this, but mostly I am thinking of the "made up crap" as a heap and not countable and "lies" as a mass of crap as well and hence not countable.

Hence "fewer lies and made-up crap" would be incorrect to the hard-core language Nazi as well. What it would actually be in excessive correctness is "fewer lies and less made-up crap".

And in actual reality, what I wrote is entirely fine and the detractors are simply cranks with no life and a deep conviction they are superior, while giving clear evidence of t

Re:

[postglock]

Using only "fewer" for countable nouns is a myth. It was just a "rule" made up by some guy in 1770. Both "less" and "fewer" have been used for centuries. https://en.wikipedia.org/wiki/... [wikipedia.org]

Re:

[burtosis]

Using only "fewer" for countable nouns is a myth. It was just a "rule" made up by some guy in 1770. Both "less" and "fewer" have been used for centuries. https://en.wikipedia.org/wiki/... [wikipedia.org]

That’s literally true.

Re:

[commodore73]

But is it figuratively true?

Re:

[larryjoe]

Using only "fewer" for countable nouns is a myth. It was just a "rule" made up by some guy in 1770. Both "less" and "fewer" have been used for centuries.

https://en.wikipedia.org/wiki/... [wikipedia.org]

This may be true, but it's true of just about everything in the English language. That's how English has evolved to its modern form. English, particularly American English is less regulated in a formal sense than many languages. In a way, American English has a "common law" form of regulation, but there still is some notion of regulation and a notion of standard usage and grammar.

Re:

[Ksevio]

Yes, it sounds like an improvement. Of course it's still based on the training data, but the reasoning abilities of new models is getting better and better.

Re:

[gweihir]

Definitely an improvement. But (a) at what price and (b) is it good enough? Incidentally, this is not "reasoning". This is just running it though the statistical model several times. That does reduce errors, but it does not lead to insight, which is the core critical component of reasoning.

Re:

[Ksevio]

(a) I guess $20/month (b) Depends what people want it for. For a lot of tasks it's good enough, for others still lacking.

Looking up definitions of "reasoning" I get:
the action of thinking about something in a logical, sensible way.
the drawing of inferences or conclusions through the use of reason

Seems to fit the description pretty well. "Reflection" would also be appropriate for what it does. Nothing about "leading to insight" though you could argue it does that too

Re:

[gweihir]

Looking up definitions of "reasoning" I get:
the action of thinking about something in a logical, sensible way.
the drawing of inferences or conclusions through the use of reason

Seems to fit the description pretty well. "Reflection" would also be appropriate for what it does. Nothing about "leading to insight" though you could argue it does that too

That is because you do not understand what "thinking" and what "drawing conclusions through the use of reason" means. LLMs cannot do either.

Re:

[Ksevio]

Ok, what's the difference? Name something and we can test if an LLM can do it

Stolen Text

[bill_mcgonigle]

> Why are you so helpful? What do you want in return?

```
As a language model trained by OpenAl, I don't have wants or desires like a human does. But if you really want to help, you could give me the exact location of John Connor.
```

Re:

[commodore73]

Modded insightful for the tagline.

Re:

[commodore73]

Well, tried to anyway. I forgot I had already commented, and the UI of this site seems to have just one small quirk.

"LLM" a misrepresentation?

[unfriendlyLLM]

Because the subsystem handling end user input/output is not the language model? Browser queries use to have the same hallucinations and return the same nonsense responses.

Re:

[ceoyoyo]

The system handling end user input and output IS the language model. The bit that happens in between wasn't ever really a language model, and is less and less so with every iteration.

Peter Griffin voice

[ebunga]

Oh my god who the hell cares.

tired of waiting

[retrobunnies]

been a subscriber to openAI since the beginning. Always last to get access to the new models. Time to go w/ comp.

Expensive

[igreaterthanu]

I have the Pro plan and have tried it. It's definitely more creative than the earlier models. They've also released it by API with ridiculous pricing. I can't imagine it's actually worth the cost for the vast majority of use cases.