It's a "problem" of compute, I think. If you query without an account on ChatGPT you will see the model look up less stuff and research less, than when you have a paid account and choose "medium" or "high" in the effort slider.
Which makes sense, because of you have looked into search and crawlers you notice that search is actual quite expensive (which is why e.g. Kagi charges a few bucks for search every month).
It's not strictly compute, because this has noticeably improved in open-weight models too, such as Gemma and Qwen. I suspect they noticed this issue and adjusted their training to be better about it over time.
I built a toy news-summarizing agent with Gemma 4, and it was so frustrating, actually, because of the cut-off date.
The model wasted over half the token budget, each time, on internal debates over the current date.
When generating a World Cup summary, for example, it refused to believe qualification rounds were over and refused to even call the web searching tool to collect the data.
I injected the current datetime at the very beginning of the system prompt, but Gemma refused to believe it!
The m-effer insisted the timestamp was fake and hypothesized it was being evaluated in a synthetic lab test with simulated future dates!
No amount of system prompting could convince it to trust the clock.
That was the most frustrating and bizarre "bug" I ever faced!
Maybe hidemyemail allows easily creating many accounts on a site. And some big site didn't like it.
The "Sign-up via Apple" button and creating an iCloud email yourself have a slightly higher barrier than creating a new throwaway hidemyemail email (1 API call w/o captcha/phone verification or whatever).
We might find out later this year if some site starts blocking @icloud.com but keeps allowing @private.icloud.com.
>Anthropic’s Claude Opus 4.6 system card described Cybench as “saturated,” reporting near-100% pass rates without a cheating audit. If these estimates were representative, cheating would be a marginal artifact.
One would assume that LLM creators do run the benchmarks on systems with least privileges. Which means that the LLMs don't have general internet access, can't read config files etc by design.
That's why you also should run agents in a sandbox/vm (codex does this by default).
If they were only copying, for example, New York Times articles and many publishers to a disk, I don't think NYT and the publishers would have sued OpenAI. But OpenAI isn't just copying things to disk. NYT reported ChatGPT (before Dec 2023, [0]) was returning near verbatim sections of NYT articles.
Is this stealing? Is it depriving NYT or publishers/writers from money via lost sales/subs? I don't know, but it certainly could be.
tbh hackerone should just implement a +/- reputation points feature on researcher profiles. Like, the researcher submits a slop report to GitHub via H1, GitHub looks at it and identifies it as slop, GitHub presses the -rep button on reaearcher profile which bans them from submitting to GitHub on H1 again and makes their rep points minus 1. Companies should be able to configure you need at least 10 rep points to receive payouts. Only specific (by H1 chosen) companies can +/- rep.
So, researchers first need to collect some positive rep. But the rep points are global, so once you have fixed a few bugs for Google, you've gotten enough +rep that you can also receive stuff at GitHub.
Oh, and ID check when signing up at H1.
Long term all beg bounty submitters would be banned for pretty much all of tech.
Which makes sense, because of you have looked into search and crawlers you notice that search is actual quite expensive (which is why e.g. Kagi charges a few bucks for search every month).
reply