Really interesting that the conversation is shifting to "LLMs are actually smart enough, we just need to build better harnesses"
Goalposts shifting?
English
Alberto Fortin
1.2K posts
Alberto Fortin
@a7fort
bootstrapped founder, fractional CTO, event organiser, music maker 🇮🇹🇬🇧 I talk tech, AI, creativity yes to nuanced, balanced takes no to bullshit hype
England Katılım Haziran 2020
339 Takip Edilen618 Takipçiler
Can we just move past this collective hallucination of AI being able to replace a senior engineer, and just make the most of what is still an incredible technological advancement?
English
Also can't help but being sceptical about the opinion of people who get featured on OpenAI's official announcement, even with the best of intentions you will be inclined to just say positive things about them
English
I'm sorry but I'm not buying it. These models are getting incrementally better but let's not throw the "AGI" word around when they still make very stupid mistakes.
I literally just had it ignore a rule in AGENTS.md - which literally has 10 lines
Pietro Schirano@skirano
One day while testing GPT-5.5, I had my first taste of AGI. We had a branch with hundreds of visual and front-end changes, plus complex refactors. At the same time, main had changed a lot too. Conflicts everywhere.
English
Alberto Fortin retweetledi
OVERRATED: running tons of agents in parallel; working on too many things at once; perpetual context-switching; opening lots of low-quality PRs that may never land.
UNDERRATED: using one or two agents at a time; focusing on the task in front of you; thinking deeply; finishing stuff; making your code works in prod.
English
@thisisjonc same, just switched last week from Claude $100/mo to Codex $100/mo
English
Codex on the $100/month plan on Extra High uses just 10% per day - that's completely reasonable.
What's not reasonable?
Any of those token-gobbling Claude regressions that Anthropic puts out like 4.7.
I'm fully sold on Codex now so long as OpenAI doesn't drop the ball.
English
I also believe this and Sam Altman kinda confirmed this, in summer 2025 nonetheless.
The raw intelligence has likely peaked and they're trying to improve the models by giving them more tools / more context / better prompts
BURKOV@burkov
For those living under a rock: LLMs stopped becoming smarter around summer 2025. Everything impressive you see since then is about finetuning them for specific tasks (mainly coding and software-tool-based task solving) and building tooling around them (such as agentic coding systems).
English
@zuess05 "when the AI stops making mistakes" this will never, ever, ever happen.
There will be fewer mistakes but unless there's a fundamental change in how LLMs work, you can never 100% trust them - for example - to have access to your production DB unassisted
English
Genuine question.
Tech companies are laying off thousands of engineers, and the ones left behind are basically just reviewing AI-generated code.
But what happens in 6 months when the AI stops making mistakes?
If your entire $200k job has been reduced to proofreading Claude's output, what exactly are they paying you for?
English
current status:
- cancelled Cursor
- cancelled Claude max
- upgraded to ChatGPT 5x
Claude is still the best for personal chats (ChatGPT style makes me vomit) but for coding I can't rely on a model that gets regularly nerfed
English
@LBacaj I'm finding the cognitive load to be much higher. Used to tackle a handful of tasks a day, now I manage+review hundreds of files a day over dozens of tasks. My brain is fried at the end of the day!
English
They’re saying AI will replace me. But tbh I can’t wait for it to replace me so I can sit back on a beach somewhere.
But so far all it’s done is make me work harder. And it feels like there is an infinite amount of ideas to explore. Claude give me a day off please.
English
Opus 4.6 has suddenly become really bad, I appreciate the speed at which Anthropic is pushing new Claude Code features but people mainly want a great model, everything else is bonus
English
Alberto Fortin retweetledi
Superwhisper's next update might be too powerful to release publicly.
The new voice model is so fast at transcription it started finishing sentences users hadn't thought of yet...
We even put it in a sandbox and it dictated its way out.
It also identified a flaw in the English language that had gone unnoticed for 600 years. Linguists have been informed.
Out of an abundance of caution, we are withholding the update until further notice.
Sincerely,
The Superwhisper Team
English
@SabatinoMasala the "out of 5" reviews system is completely broken and it needs to be rehauled for it to be meaningful. People avoid a 3.9 restaurant, but if you said your meal was 7.8/10 it would be a compliment. None of it makes sense!
3.0 should mean 'satisfactory', not 'food poisoning'
English
A thing I’ve been noticing in big cities is that restaurant reviews are super inflated. 4.8 on 1k+ reviews is at most a 3/5 experience 😅 Fake reviews? Incentivised reviews? In any case, Google reviews are not to be trusted anymore 😔
English
@marcelpociot @shadcn @getpolyscope omg this is exactly what I was looking for. will test it out this week
English
@shadcn That's why I built @getpolyscope 😅
We use copy on write clones instead, so you get all dependencies of your repositories basically for free (and super fast)
getpolyscope.com
English
hey @sama did you actually almost buy a Business Yearly plan on tokchart.com in July last year?
you might not need it but I can definitely work out a discount for you..
English
@nikpolale great work on sequenzy, I just signed up and I love the AI-first approach.
I'm having an issue with sequences though, I can't save the sequence when I put an if/else statement, I always get an error like this
English
@iannuttall 400 total saves on these tweets ahahaha idiocracy was truly a documentary
English
I also feel like devs have 0 loyalty and are so quick at switching tools, so I don't think Codex has "won" anything permanent.
It used to take years and a lot of convincing to switch IDE, now you go from Cursor to Claude Code to Codex in the span of a few monhts
English
I just realised I haven't used Claude Code in a while.
Things I prefer about Codex:
- clean UI with threads organised by repo
- very generous tokens allowance
Mitchell Hashimoto@mitchellh
I know this is pretty well established at this point, but Codex 5.3 is a much more effective model than Opus 4.6. I went back and forth on both for a bit, but haven’t touched Opus at all now for a full week. First model to get me off of Opus… ever. Good job Codex team.
English