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

GPT-5.2

openai.com

Post reply on HN
Print view View original

531–540 of 1001 posts

Re: GPT-5.2

#531

Post by gcr »

sundarurfriend wrote: post #363
solarkraft wrote: post #306

Earlier quoted context omitted.

> Only downsides are in the polish department What an understatement. It has me thinking „man, fuck this“ on the daily. Just today it spontaneously lost an entire 20-30 minutes long thread and it was far from the first time. It basically does it any time you interrupt it in any way. It’s straight up data loss. It’s kind of a typical Google product in that it feels more like a tech demo than a product. It has theoreti…

ChatGPT web UI was also like this for the longest time, until a few months ago: all sorts of random UI bugs leading either to data loss or misleading UI state. Interrupting still is very flaky there too. And on the mobile app, if you move away from the app while it's taking time to think, its state would somehow desync from the actual backend thinking state, and get stuck randomly; sometimes restarting the app fixes…

i basically can't use the ChatGPT app on the subway for these reasons. the moment the websocket connection drops, i have to edit my last message and resubmit it unchanged.

it's like the client, not the server, is responsible for writing to my conversation history or something

Re: GPT-5.2

#532

Post by astrange »

gs17 wrote: post #383
lxgr wrote: post #313

Earlier quoted context omitted.

Interesting, I had the opposite experience. 5.0 "Thinking" was better than 5.1, but Gemini 3 Pro seems worse than either for web search use cases. It's hallucinating at pretty alarming rates (including making up sources it never actually accessed) for a late 2025 model. Opus 4.5 has been a step above both for me, but the usage limits are the worst of the three. I'm seriously considering multiple parallel subscription…

I've had the same experience with search, especially with it hallucinating results instead of actually finding them. It's really frustrating that you can't force a more in-depth search from the model run by the company most famous for a search engine.

Try the same question in deep research mode.

Re: GPT-5.2

#533

Post by dominotw »

free652 wrote: post #525
radicaldreamer wrote: post #337

Earlier quoted context omitted.

Google’s standard problem is that they don’t even use their own products. Their Pixel and Android team rocks iPhones on the daily, for example.

You cant buy an iPhone without a director approval. And it's like 3 gen behind as well. So no, they don't use iPhones.

you have to get premission from director for your presonal phone? wtf

Re: GPT-5.2

#534

Post by ummonk »

free652 wrote: post #525
radicaldreamer wrote: post #337

Earlier quoted context omitted.

Google’s standard problem is that they don’t even use their own products. Their Pixel and Android team rocks iPhones on the daily, for example.

You cant buy an iPhone without a director approval. And it's like 3 gen behind as well. So no, they don't use iPhones.

Google tells its employees what products they're allowed to buy for personal use?

Re: GPT-5.2

#535

Post by astrange »

redhed wrote: post #434
noname120 wrote: post #392

Earlier quoted context omitted.

Problem is that by default ChatGPT has the “Reference chat history” option enabled in the Memory options. This causes any previous conversation to leak into the current one. Just creating a new conversation is not enough, you also need to disable that option.

This is also the default in Gemini pretty sure, at least I remember turning it off. Make's no sense to me why this is the default.

Mostly because they built the feature and so that implicitly means they think it's cool.

I recommend turning it off because it makes the models way more sycophantic and can drive them (or you) insane.

Re: GPT-5.2

#536

Post by piskov »

Kim_Bruning wrote: post #518

I'm continuously surprised that some people get good results out of GPT models. They sort of fail on my personal benchmarks for me. Maybe GPT needs a different approach to prompting? (as compared to eg Claude, Gemini, or Kimi)

They are all gpt as in generative pre-trained transformer

Re: GPT-5.2

#537

Post by astrange »

ssl-3 wrote: post #100
fouronnes3 wrote: post #20

Earlier quoted context omitted.

I want to read a short scify story set in 2150 about how, mysteriously, no one has been able to train a better LLM for 125 years. The binary weights are studied with unbelievably advanced quantum computers but no one can really train a new AI from scratch. This starts cults, wars and legends and ultimately (by the third book) leads to the main protagonist learning to code by hand, something that no human left alive s…

Sounds good. Might sell better with the protagonist learning iron age leatherworking, with hides tanned from cows that were grown within earshot, as part of a process of finding the real root of the reason for why any of us ever came to be in the first place. This realization process culminates in the formation of a global, unified steampunk BDSM movement and a wealth of new diseases, and then: Zombies. (That's the e…

This is somewhat similar to a Piers Anthony series that I suspect noone has ever read except for me.

What was with that guy anyway.

Re: GPT-5.2

#538

Post by gcr »

free652 wrote: post #525
radicaldreamer wrote: post #337

Earlier quoted context omitted.

Google’s standard problem is that they don’t even use their own products. Their Pixel and Android team rocks iPhones on the daily, for example.

You cant buy an iPhone without a director approval. And it's like 3 gen behind as well. So no, they don't use iPhones.

lots of googlers use BYOD iPhones and the corp suite for this use case is fairly well-supported

Re: GPT-5.2

#539

Post by astrange »

rockinghigh wrote: post #219
Wowfunhappy wrote: post #17

Earlier quoted context omitted.

I thought whenever the knowledge cutoff increased that meant they’d trained a new model, I guess that’s completely wrong?

They add new data to the existing base model via continuous pre-training. You save on pre-training, the next token prediction task, but still have to re-run mid and post training stages like context length extension, supervised fine tuning, reinforcement learning, safety alignment ...

Continuous pretraining has issues because it starts forgetting the older stuff. There is some research into other approaches.

Re: GPT-5.2

#540

Post by dcre »

vintermann wrote: post #353
onraglanroad wrote: post #333

I suppose this is as good a place as any to mention this. I've now met two different devs who complained about the weird responses from their LLM of choice, and it turned out they were using a single session for everything. From recipes for the night, presents for the wife and then into programming issues the next day. Don't do that. The whole context is sent on queries to the LLM, so start a new chat for each topic.…

It's not at all obvious where to drop the context, though. Maybe it helps to have similar tasks in the context, maybe not. It did really, shockingly well on a historical HTR task I gave it, so I gave it another one, in some ways an easier one... Thought it wouldn't hurt to have text in a similar style in the context. But then it suddenly did very poorly. Incidentally, one of the reasons I haven't gotten much into sub…

The models you interact with through the API (as opposed to chat UIs) are held stable and let you specify reasoning effort, so if you use a client that takes API keys, you might be able to solve both of those problems.
Post reply on HN