Chris Prosser

Software developer / ML Team Lead

[Chris Prosser @ProsserCJ]

@thsottiaux Better judgement on when to ask for user guidance vs push on. I'm hitting examples where it can't do what we outlined (e.g. some external constraint), finds a workaround that isn't what I wanted and spends lots of tokens planning in detail or building rather than checking in.

[Chris Prosser @ProsserCJ]

@claudeai This is cool, but on naming Mythos sounds so much more epic than Fable!

[Chris Prosser @ProsserCJ]

@thsottiaux I think benchmarks do have signal for real usefulness but not the full picture. I try the new OpenAI and Anthropics models by default because of track record and others if people I trust recommend or if they do very well on an index like artificial analysis.

[Chris Prosser @ProsserCJ]

@sama Understanding intent. I recently switched from Claude to mostly Codex due to rate limits and was pleasantly surprised by how strong GPT-5.5 is on well-defined tasks. Where it still lags Claude IMO is messy/iterative work: knowing when to ask vs act, and getting what I mean.

[Chris Prosser @ProsserCJ]

@dhh I'm going to have so much fun with this 😆

[Chris Prosser @ProsserCJ]

@waitbutwhy This isn't a simple question of save yourself or others as a lot of people in the comments seem to think. This is a coordination problem, if everyone voted the same colour everyone is fine. The trouble is trying to guess what everyone will vote.

[Chris Prosser @ProsserCJ]

@bcherny The funny thing is their Claude probably affirmed their brilliance and helped them draft the email 😆

[Chris Prosser @ProsserCJ]

@josefbender_ TBH I would rather some of these were 10 mins, I like some of Theo's takes, but I'm not spending an hour a day listening to him, so skip most videos

[Chris Prosser @ProsserCJ]

@theo @bcherny please sort this out. I love Claude code, but want choice over how we use it!

[Chris Prosser @ProsserCJ]

@sama Very disappointed

[Evan Hubinger @EvanHub]

We may yet fail to rise to all the challenges posed by transformative AI. But it is worth celebrating that when it mattered most and we were asked to compromise the most basic principles of liberty, we said no. I hope others will join. https://t.co/lSJnCluAPn

[Chris Prosser @ProsserCJ]

Anyone else noticing issues with the new Opus 4.6 model this morning? I'm using it in Claude code and finding it's making more incorrect assumptions - confidently. I'm having to watch the reasoning, spot errors and prove it's wrong via web search or screenshots. @AnthropicAI

[Chris Prosser @ProsserCJ]

@sama @AnthropicAI are raising concerns that we feel as users. They are exaggerated scenarios to add humour sure. But, it's easy to make promises about principles and separation now, but the track record doesn't show these reassurances standing over time e.g. non profit oversight

[Chris Prosser @ProsserCJ]

@bcherny @karpathy Each developer opening 20+ PRs a day sounds like hell to me - does a member of staff review all of these or do you rely on Claude for this too? I have found, similar to @karpathy that I need to keep a close eye on the generated code for real projects.

[Chris Prosser @ProsserCJ]

@milan_milanovic I agree, for most real work I still use the IDE, check changes at small increments and edit or steer. This limits the speed up, but I need to understand the code generated on important projects. On more experimental work, I like the terminal.

[Chris Prosser @ProsserCJ]

@eevblog I like Omarchy, but if you don't want to learn key combos and want something that will be more familiar to a Mac or Windows like environment then I would recommend Ubuntu.

[Chris Prosser @ProsserCJ]

@AmandaAskell 100% - as if it's a binary thing.

[Programmer Humor @PR0GRAMMERHUM0R]

theFourStagesOfACodeReview https://t.co/aupDt8c4nu

[Chris Prosser @ProsserCJ]

@OpenAI @OpenAI - I'm having issues with really basic reasoning errors / instruction following ability on the latest GPT 5.2. In case it's helpful here is an example where it was painful to get the model to follow simple instructions: LLM release comparison https://t.co/CODfeg25JF

[Chris Prosser @ProsserCJ]

@ash_twtz @kimmonismus Opus is far better IMO