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

GPT-4.5

openai.com

Post reply on HN
Print view View original

561–570 of 1001 posts

Re: GPT-4.5

#561

Post by maeil »

tmaly wrote: post #95
Topfi wrote: post #24

Considering both this blog post and the livestream demos, I am underwhelmed. Having just finished the stream, I had a real "was that all" moment, which on one hand shows how spoiled I've gotten by new models impressing me, but on another feels like OpenAI really struggles to stay ahead of their competitors. What has been shown feels like it could be achieved using a custom system prompt on older versions of OpenAIs m…

rethinking your comment "was that all" I am listening to the stream now and had a thought. Most of the new models that have come out in the past few weeks have been great at coding and logical reasoning. But 4o has been better at creative writing. I am wondering if 4.5 is going to be even better at creative writing than 4o.

> But 4o has been better at creative writing

In what way? I find the opposite, 4o's output has a very strong AI vibe, much moreso than competitors like Claude and Gemini. You can immediately tell, and instructing it to write differently (except for obvious caricatures like "Write like Gen Z") doesn't seem to help.

Re: GPT-4.5

#562

Post by JohnnyMarcone »

goatlover wrote: post #211
wavemode wrote: post #171

Earlier quoted context omitted.

This has been my suspicion for a long time - OpenAI have indeed been working on "GPT5", but training and running it is proving so expensive (and its actual reasoning abilities only marginally stronger than GPT4) that there's just no market for it. It points to an overall plateau being reached in the performance of the transformer architecture.

Certainly hope so. The tech billionaires are little to excited to achieve AGI and replace the workforce.

I feel like this period has shown that we're not quite ready for a machine god. We'll see if RL hits a wall as well.

Re: GPT-4.5

#563

Post by dpe82 »

dzjkb wrote: post #508
planetafro wrote: post #394

Earlier quoted context omitted.

I think search and chat are decent products as well. I am a Google subscriber and I just use Gemini as a replacement for search without ads. To me, this movement accelerated paid search in an unexpected way. I know the detractors will cry "hallucinations" and the ilk. I would counter with an argument about the state of the current web besieged by ads and misinformation. If people carry a reasonable amount of skeptici…

speaking of search without ads, I wholeheartedly recommend https://kagi.com

I'll second this. Kagi is really impressive and ad-free is a nice change.

Re: GPT-4.5

#564

Post by JohnnyMarcone »

jdprgm wrote: post #84
zaptrem wrote: post #26

GPT 4.5 pricing is insane: Price Input: $75.00 / 1M tokens Cached input: $37.50 / 1M tokens Output: $150.00 / 1M tokens GPT 4o pricing for comparison: Price Input: $2.50 / 1M tokens Cached input: $1.25 / 1M tokens Output: $10.00 / 1M tokens It sounds like it's so expensive and the difference in usefulness is so lacking(?) they're not even gonna keep serving it in the API for long: > GPT‑4.5 is a very large and comput…

If it really costs them 30x more surely they must plan on putting pretty significant usage limits on any rollout to the Plus tier and if that is the case i'm not sure what the point is considering it seems primarily a replacement/upgrade for 4o. The cognitive overhead of choosing between what will be 6 different models now on chatGPT and trying to map whether a query is "worth" using a certain model and worrying abou…

To be fair their roadmap states that gpt-5 will unify everything into one model in "months".

Re: GPT-4.5

#565

Post by tmpz22 »

hombre_fatal wrote: post #547

Cathartic moment over.

I haven't had the same experience. Here are some of the significant issues when using o1 or claude 3.7 with vscode copilot:

* Very wreckless in pulling in third party libraries - often pulling in older versions including packages that trigger vulnerability warnings in package managers like npm. Imagine a student or junior developer falling into this trap.

* Very wreckless around data security. For example in an established project it re-configured sqlite3 (python lib) to disable checks for concurrent write liabilities in sqlite. This would corrupt data in a variety of scenarios.

* It sometimes is very slow to apply minor edits, taking about 2 - 5 minutes to output its changes. I've noticed when it takes this long it also usually breaks the file in subtle ways, including attaching random characters to a string literal which I very much did not want to change.

* Very bad when working with concurrency. While this is a hard thing in general, introducing subtle concurrency bugs into a codebase is not good.

* By far is the false sense of security it gives you. Its close enough to being right that a constant incentive exists to just yeet the code completions without diligent review. This is really really concerning as many organizations will yeet this, as I imagine executives are currently the world over.

Honestly I think a lot of people are captured by a small sample size of initial impressions, and while I believe you in that you've found value for use cases - in aggregate I think it is a honeymoon phase that wears off with every-day use.

Re: GPT-4.5

#566

Post by abrichr »

apwell23 wrote: post #554
hombre_fatal wrote: post #547

Cathartic moment over.

doesn't feel like to me. I try using copilot on my scala projects and it always comes up with something useless that doesn't even compile. I am currently just using it as easy google search.

Have you tried copying the compilation errors back into the prompt? In my experience eventually the result is correct. If not then I shrink the surface area that the model is touching and try again.

Re: GPT-4.5

#567

Post by osigurdson »

zaptrem wrote: post #26

GPT 4.5 pricing is insane: Price Input: $75.00 / 1M tokens Cached input: $37.50 / 1M tokens Output: $150.00 / 1M tokens GPT 4o pricing for comparison: Price Input: $2.50 / 1M tokens Cached input: $1.25 / 1M tokens Output: $10.00 / 1M tokens It sounds like it's so expensive and the difference in usefulness is so lacking(?) they're not even gonna keep serving it in the API for long: > GPT‑4.5 is a very large and comput…

I don't understand the pricing for cached tokens. It seems rather high for looking up something in a cache.

Re: GPT-4.5

#568

Post by wordpad25 »

stevage wrote: post #528
simonw wrote: post #342

I got gpt-4.5-preview to summarize this discussion thread so far (at 324 comments): hn-summary.sh 43197872 -m gpt-4.5-preview Using this script: https://til.simonwillison.net/llms/claude-hacker-news-themes... Here's the result: https://gist.github.com/simonw/5e9f5e94ac8840f698c280293d399... It took 25797 input tokens and 1225 input tokens, for a total cost (calculated using https://tools.simonwillison.net/llm-prices…

Huh. Disregarding the 4.5-specific bit here, a browser extension or possibly website that did this in general could be really useful. Maybe even something that just noticed whenever you visited a site that had had significant HN discussion in the past, then let you trigger a summary.

there are literally hundreds of extensions and sites that do this

the problem is that they are competing each other into the ground hence they go unmaintained very quickly

getrecall.ai has been the most mature so far

Re: GPT-4.5

#569

Post by kneegerman »

dgfitz wrote: post #380

@sama, LLMs aren't going to create AGI. I realize you need to generate cash flow, this isn't the play. Sincerely, Me

you misunderstand, the business model is extracting cash from Qatar et al

Re: GPT-4.5

#570

Post by chippiewill »

sebastiennight wrote: post #154
eightysixfour wrote: post #141

Earlier quoted context omitted.

From Sam's twitter: > After that, a top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks. > In both ChatGPT and our API, we will release GPT-5 as a system that integrates a lot of our technology, including o3. We will no longer ship o3 as a standalone model. Yo…

Ah, great point. Yes, the wording here would imply that they're basically planning on building scaffolding around multiple models instead of having one more capable Swiss Army Knife model. I would feel a bit bummed if GPT-5 turned out not to be a model, but rather a "product".

Which is intriguing in a way, because the momentum I've seen across AI over the past decade has been increasing amounts of "end-to-end"
Post reply on HN