Sabitlenmiş Tweet
The secret to success: they WON'T tell you:
> sit down
> become very still
> try to do absolutely nothing for as long as possible
> keep doing that until nirvana is reached
> escape the endless cycle of death and rebirth
English
Gabriel Garrett
12.5K posts
Gabriel Garrett
@GabGarrett
10,000x AI Engineer. Chief Singularity Officer member of the permanent overclass Writing on AGI Futurism / Mindfulness / Panpsychism @ https://t.co/xoTtQR5NIm
Panhandle, FL Katılım August 2018
1.1K Takip Edilen2.3K Takipçiler
2026 Yıllık Özeti
@GabGarrett hesabının Twitter yılını gör
@boneGPT same issue in Pensacola. they are in total tiktok psychosis over flock cameras and datacenters
English
There is petition against datacenters in my community of Florida boomers. They are talking about water and the books.
These people believe they are well informed. There are comments from the right calling it deep state.
When morons on both sides agree, we're in trouble.
English
gmail has completely nuked all search functionality in their app. completely insane.
English
Gabriel Garrett retweetledi
@sheherenow_ Some people seem to literally be stuck in March of 2020 as though they are models with a knowledge cutoff date
English
you honestly gotta respect the level of cope and the ability to never ever admit he was wrong. truly the GOAT
English
Gabriel Garrett retweetledi
@poiThePoi @rahulj51 As we lose our will and practice for digging into code, we'll end up just begging and pleading agents to fix things, like shamans dancing for rain.
English
@johnennis I think this is highly dependent on how soon we get functional mass producible humanoid robots
English
I think two things are true at the same time:
1. We are in the singularity
2. It won’t be as big deal as everyone thinks
There will be a period of adjustment and then life will go on
English
@yacineMTB even if you tried to tell them, they wouldn’t understand, or even believe
English
Gabriel Garrett retweetledi
Gabriel Garrett retweetledi
Shakespeare invented hundreds of words; words that we all still use every day. He needed a denser language.
It’s possible that as LLM-speak becomes denser that our language will change again to adopt the abstraction offered by LLM verbiage
merlin@merlindru
@MarcJSchmidt the busted RL is also why its so smart though. i hypothesize that LLMs just really really really want a denser language. english is too limiting, so they invent terms that hold more information. why use a word that expresses one thing if you could use another word to express 3?
English
It is actually crazy how hard it is to convince people to get a paid AI subscription. You can go to insane lengths to show them how many hours they would save and they won’t fork up even the cost of a couple of beers
Bren@BrenBuilds
@banteg Uncle was reconsiling a spreadsheet for his business and free llm roulette in the browser. Showed him my laptop with codex app take control and fill out Google sheets instantly. Showed excel plugin. Told him I pay for the service. He's still playing free llm roulette
English
English
@gdb Why is it not contacting the other person’s chatgpt instead of the person?
English
at openai, many people hook their chatgpt up to slack.
people really don't like when a coworker's chatgpt contacts them asking for help with a task, even when they'd be perfectly happy doing that same work if asked by that coworker.
reinforces how much people care about human relationships and helping each other, and want AI to give time back — or enhance time together — rather than become a layer separating people.
English
@nickscamara_ but what about the OCR part? Still often needed
English
we built pdf-inspector so agents can process PDFs without waiting on OCR. it classifies any PDF in ~20ms and extracts clean markdown locally
→ 200 PDFs processed in 2.8s
→ top quality in extracting tables + graphs
→ built in rust
→ open source
github.com/firecrawl/pdf-…
English
intelligence is actually coming down to the cost of electricity
DeepSeek@deepseek_ai
🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex! Check out the configuration details in our official API docs: api-docs.deepseek.com/quick_start/ag…
English
@advaitpaliwal @doodlestein sub-management could be his next paid SaaS or open-source project, I’d probably sign up for that
English
@doodlestein @GabGarrett How are you managing all these subs
English
The real software factory is literally what I am doing day in, day out.
And it's not 10x, it's not even 100x, it's more like 1000x. Lots of authorities on the subject, but where are the repos proving they're doing it?
It reminds me of Ronnie Coleman's line, "Everybody wants to be a bodybuilder but nobody wants to lift these heavy ass weights."
If you want to have a software factory, then you need to show you can lift the heavy ass weights: the commits, the projects, the stars, the usage, the value creation, the cumulative flywheel.
And I will put mine up against literally anyone in the world:
github.com/Dicklesworthst…
Hamid Dadkhah@hamidships
the real "software factory" isn’t everyone using coding agents. that’s the 2x. the 10x is a different thing entirely. the loop starts closing on its own. 2x = every engineer using agents, ~100% AI-generated code. soon, that’ll be table stakes. you made the same loop faster, but the bottleneck is still human attention at every gate. 10x = the loop self validates. agents stop waiting on you. but it only works if the scaffolding is real: comprehensive CI, evals that actually catch regressions, and end-to-end test environments agents spin up to check their own work. to be clear, 10x is not "fewer humans." it is humans on the decisions that carry real risk and off the rote gates. we still need to hire exceptional engineers, their judgment is the scarce input now, so you spend it where it actually counts. but here is what breaks traditional SaaS instincts: building agents is divergent. fix one thing and 10 new things need evaluating. converging on that locally is a trap. so ship the 80/20 fast, but through controlled early access: opt-in design partners, clear guardrails, tiered rollout. never unfinished work dumped on people who did not sign up for it. then close the loop. feedback-loop agents take that signal, implement it, merge it, validate post-deploy, continuously. and because CI/CD is fast, the gap between "customer said X" and "X is live and validated" collapses. coding at the speed of thought. give agents the same context a human uses to make the call, user feedback, docs, prod monitoring, ephemeral test clusters. encode your review standards as rules the reviewer agent reads. auto-merge the low-risk stuff classified by your reviewer agent. reserve human eyes for billing, auth, the paths where a mistake is expensive. humans do not disappear, they concentrate. incidents stay human-led, business impact's too high to hand off, though agents now dig datadog + telemetry + code and hand you impact + root cause + mitigation in minutes. and someone guards the critical paths: as agentic volume climbs you risk losing deep understanding of your own codebase. the fix isn't to slow down — it is to lock down what's expensive to get wrong. Not all code carries equal risk. at Ramp, this is Inspect, our internal coding agent: deep Ramp context, MCP integrations, CI visibility, and self-validation that boots the UI and tests real workflows. On a recent UI change, it deployed a preview, authenticated as a fixture user, validated desktop and mobile behavior, and attached visual evidence to the PR. the metric I actually care about: fewest human touches per safely merged PR. not lines, not PR count. how much can the loop close on its own without breaking the things that matter and do it super fast? that's the edge.
English
I don’t think this is true. At my work, we had a really messed up streaming message parser that had grown into 20,000 lines of broken patch attempts and tests.
I told fable “this is really messed up and we’ve tried to fix this many times. Can you make this actually work right?” It dug through the whole commit history, deleted 20k lines, and added 1,500. It worked.
English
(sorry for the silly post just some thoughts)
so after working with Fable 24/7 for several weeks now, I now genuinely believe that, indeed, the only and one last thing between us and AGI is erasure. I can't put in words how strong this belief is forming in me. I really believe so. not architecture, not compute, not datasets. every capability that the AI currently lacks (continual learning, infinite context, long term work, etc.) is root-caused by its inability to erase, and they will all emerge naturally from a well designed erasure centric train, just like so many capabilities emerged from the reasoning paradigm
I really wish I could do something about this but I'm not in a big lab so, meanwhile, I've been thinking in ways to patch that externally. clearly begging the AI to "remove, re-derive" does NOT work, because it has a massive bias to keeping what is there. an example from today: I spent an hour *begging* Fable to simplify Bend2's parser as much as it could, and I was paying attention to one specific function, a completely moronic backtracker that should NEVER be in a well written parser. yet, it was never removed. it was untouched, survived several cleanup turns, no matter how I prompted Fable, how much I asked it to aggressively prune the code. it is simply incapable of refactoring competently because Fable just... adds, patches, and keeps
but then, I realized: it cannot keep something it doesn't see!
so, what if I do this instead?
1. ask Fable to select some code and describe what it is *for*
2. then, a script mechanically removes that code
3. the script asks Fable to fill it again, without seeing the original
4. ...?
5. AI capable of improving long term code quality
basically I think I could perhaps compensate Fable's utter inability to erase stuff by just... deleting that stuff myself. and I then ask it to re-fill. I don't need to beg, I don't need to explain how to redesign from first principles, or meaning of erasure. I can just... literally, programmatically erase it myself. it cannot preserve stupid shit it can't see
sorry this post is as dumb as it seems, but it's not stupid if it works, right? I'm exited to try it, at least.
English
@doodlestein there is still a valuable lesson in that! the concept of a job itself is being simplified and abstracted: receive event, take action, run in a loop, and you can apply that to automating any job
English
@GabGarrett I mean, it helps that my clients are big hedge funds. And I can only get those clients because I did their jobs at big hedge funds for 10+ years, too.
English
@GabGarrett karşılaştırmaları
@bonegpt vs @gabgarrett @gabgarrett vs @sheherenow_ @benglickenhaus vs @gabgarrett @gabgarrett vs @poithepoi @gabgarrett vs @rahulj51 @gabgarrett vs @johnennis
Kendi karşılaştırmanı oluştur 
