Adam

Letting Agents run wild at a Hedge Fund

[Adam @Adamfg97]

@theo @anshnanda Any plans to bring similar usage tracking to T3 Code?

[Adam @Adamfg97]

@dabit3 Please can I have access? I am building the future of work assessments for knowledge work candidates applying to AI native companies

[Adam @Adamfg97]

@theo @jullerino It's almost 7am theo, please get some sleep

[Adam @Adamfg97]

@dreamofabear When will this be available on Enterprise?

[Adam @Adamfg97]

Hmmm 🤔

Pretty sure this is just an extremely ironic example of how even frontier models hallucinate, but it is werid as we go sol -> terra -> luna the reported model numbers go down. Maybe OpenAI is just very eco friendly and likes recycling?

[Adam @Adamfg97]

@thsottiaux Why can't we save these resets for later like normal? Pls Tibo :(

[Adam @Adamfg97]

@thsottiaux Please make sure the usage tracking for GPT 5.6 is very broken so you end up handing out a load of resets

[Adam @Adamfg97]

@theo What are the pros and cons of it vs Fable? What tasks is Fable better at? What tasks is Sol better at?

[Adam @Adamfg97]

@thsottiaux Codex Desktop on Windows with the agent operating in WSL is painfully slow. It takes 5+ seconds to spawn a new thread (presumably sandbox setup) and time to first token and TPS is noticeably slower compared to GPT 5.5 Thinking in ChatGPT

[Adam @Adamfg97]

@thsottiaux The only way to use a browser with Codex CLI is the playwright MCP. If the model takes a screenshot the base 64 data is attached to the payload and a handful of these are enough for thr json payload to exceed 32mb and corrupt the session.

[Adam @Adamfg97]

@theo Ok in the browser is a much better experience than I was expecting. Thanks for the reply and thanks @jullerino for all the high-quality engineering

[Adam @Adamfg97]

@trq212 @trq212 if everything is working correctly this would mean a pro user with 20x less usage could only have this single conversation every 5 hours

[Adam @Adamfg97]

@trq212 This conversation used up 5% of my 5 hourly max x20 usage limit. It started after 8pm GMT: https://t.co/YelbwtI5dA. I watched the usage meter grow from 6 to 11%. I did not do any other chats in this time and verified no claude code sessions were running. Surely this cant be?

[Adam @Adamfg97]

@ListenLabs @ListenLabs this challenge is at least pseudo-random as far as I can tell. The top of the board has good luck, so the only way to win even with an optimal strat is luck. Can I hammer your API with 3,050 concurrent requests for the next 10 days? If not there is no point trying.

[Adam @Adamfg97]

@OpenAIDevs Do you know when the Codex web interface will be updated to use GPT5 as its base model? I believe Codex CLI has already been updated

[Adam @Adamfg97]

@c2c_Rail Why do you keep sending 4 carriage services from fenchurch Street to pitsea during the evening rush hour? This is the second day in a row where not everyone can get on at West ham because its a 4 carriage train instead of 8

[Adam @Adamfg97]

@CheeseMaster_Pr Your new game, Spirittea, looks great! Unfortunately, it's being used by scammers to spread malware on Discord. Perhaps you could get the blog they use to distribute the malware taken down via a copyright strike? https://t.co/9RJ9AuEpUo

[Adam @Adamfg97]

@SamsungHelpUK I've talked to your online shop support on the phone, and they've informed me it's not possible to claim the 25% off the galaxy watch 5 when pre-ordering the fold 4 as a new member of the Samsung Upgrade Programme. There is no mention of this anywhere on the site.

[Adam @Adamfg97]

@Tesla @elonmusk Flying Through Giga Berlin is a great video! https://t.co/tnvGNhptws. The music is great, can we please get a download for it?