Nexic Turbo

@thsottiaux … that it couldn’t access its browser for whatever reason, and would tell me itself that the way to fix it was to close/ reopen the codex app. That would work sometimes, but other times the issue would arise again. I would love to be an early tester of future/ faster browser us

@thsottiaux I used Work to autonomously apply for jobs. I had it apply for around 20 jobs yesterday. Another thing I would like improved/ fixed is browser use on desktop. It can be good, but there were times when it was trying to work on those jobs, but it would come back to me and say…

If you use ChatGPT Work. Looking to assemble a group of 50-100 to give early access to features and discuss how we can make it better. Send me a DM, mentioning ChatGPT Work, with how you use it and what you think we should fix and I’ll assemble a group in the next 48 hours. Not looking for developers this time!

@thsottiaux I would love it if, for example, when Full Access is turned on, when you give it EXPLICIT permission to do something (as long as it doesn’t violate policy of course), it actually does it.

@thsottiaux I use 5.6 sol Work to autonomously apply for jobs, and just yesterday had it apply to over 20 jobs. I gave it all my info/ credentials and had it craft a very good resume for me. I have 2 issues. One is that even when I give Work explicit permission to do certain things, and it…

@thsottiaux … agrees (as the things I give it permission to do aren’t violating policy), it will just forget that I gave it permission and told it to NOT stop to ask me for such questions. This has happened repeatedly and caused me to constantly have to give it permission AGAIN.

@goodside GPT 5.6 Sol Max, after working autonomously for 17 hours straight with NO internet access allowed, has solved it completely. codex://threads/019f6dd6-00c9-7993-b5cb-899832654949

ChatGPT 5.6 Sol Pro tries to solve a clue-free crossword of all 1,025 Pokémon (made by Fable 5 Max). Five attempts, memory off. All tries returned partial solutions at most: 1) 45 answers in 28m 2) 145 answers in 33m 3) 40 answers in 31m 4-5) Timeout/error after ~30m

@TheeShallNotSin @OverfitForTruth @goodside Yes I’m doing this now. I gave I to GPT 5.6 Sol Max and it’s been working on it for over 7 hours and still going.

@TurboNexic @OverfitForTruth @goodside If you run into the error of it timing out give the problem to codex sol

@splitbycomma Google. It says AI assists engineers across code reviews, testing, and migrations, with an estimated 10% increase in engineering velocity. Meta and Microsoft also publicly document internal AI coding workflows. That’s not mentioning the huge gains Anthropic and Open Ai report.

Name ONE highly valued company that allows its software engineers to regularly use AI. You can’t because there are none. If “vibe coding” was so good, it would have been adopted by industry leaders already.

@qCosmoCore No it isnt. You probably haven’t used frontier models like Fable or GPT 5.6 Sol. When you do, let me know!

@kayleelove2all @Lunaria_40 “Countless holes” only works if we abandon the riddle’s everyday scale; thread gaps aren’t normally counted as holes in a shirt. A sleeve opening is still a hole even if it’s essential to the garment—being necessary doesn’t stop it from being a hole.

@Lunaria_40 Technically, there are countless holes because there are gaps between the threads. However, the definition of a hole matters. Is the opening for a sleeve considered a hole? I'd say no, because if you remove it, it won't be a shirt anymore.

@maryo1952m @Lunaria_40 18 uses inconsistent counting. The 6 tears go through both front and back, making 12 holes. Add the shirt’s 4 normal openings—neck, 2 sleeves and bottom—for 16 total. The front and back edges of one opening aren’t separate holes; that logic would make 20, not 18.

@Lunaria_40 18 - hole in front and back of neck, sleeves and shirt holes!

@Lunaria_40 16 holes: 6 cutouts go through both the front and back (12), plus the neck, bottom, and two sleeves (4).

@OverfitForTruth @goodside I’m retrying now and telling it that it’s not allowed to look for an answer key or use X

@TurboNexic @goodside Problem is that it’ll most likely search and find that post. You either have to create a novel puzzle or ensure from thinking traces that it didn’t use this.

@goodside Ok fine I’ll try again with no link.

@TurboNexic No, it just found this very Twitter thread which contains the answer key. It even links to it in the response.

@goodside Full chat: chatgpt.com/share/6a5800b0…

@goodside 5.6 Pro solved it completely… chatgpt.com/s/t_6a58004f9c…

The primary reason why I love GPT 5.6 Sol is its tenacity. You give this model a goal, and it will try everything to get it done. Sometimes it isn’t perfect and may do things you don’t like, but as this is perfected over time, it will be one of the most important qualities.

@mweinbach But we didnt. Labs apart from Open Ai and Anthropic are barely releasing models at 5.5/ Opus 4.8 levels of capabilities (they are catching up to Open AI’s/ Anthropic’s last gen models…) 5.6 and Fable are the top dogs right now.

Funny how we went from like 3 labs in the race to 5 overnight. Basically 2 days and nearly doubled the models, with all being cheaper. Wild times ahead.

@Clix Give it to me, please. Thanks.

If USA wins this game I'll give one person who likes this tweet a PC