Fausto Albers

@lydiahallie Published this last wknd! github.com/Galbaz1/video-…

Excited to announce Claude for Open Source ❤️ We're giving 6 months of free Claude Max 20x to open source maintainers and core contributors. If you maintain a popular project or contribute across open source, please apply! claude.com/contact-sales/…

All we really now have to build are the tools. Agents do, and will do more before it’s 2027, understand how to use them. Almost everyone will be building code by the end of this year. And if you’re not, chances are you are missing out. I had AI build this plugin, one weekend.

@sama @thorstenball Speed!

@thorstenball what's been working well, and what would you like to see us improve in 5.3?

man, GPT-5.2 is very good.

I studied to be a sociologist, became a restaurant owner, and then transitioned to AI research. I now lead GenAI R&D at AUAS Industrial Digital Twins Lab. What I have learned contradicts much of what people assume about AI and coding. Essay -> wonderwhy.ai

@FU_joehudson Evolutionary psychology suggests our minds are modular—more like a committee of specialized drives shaped by ancestral pressures. Ideally, they integrate, but selectively accepting or rejecting parts is often key to adaptation.

26. People cannot be split up into parts you accept and parts you reject. A person is a whole.

How to be more emotionally intelligent (without trying so hard) 🧵 for @threadapalooza

@superwhisper @superwhisper is so extremely useful. I use it to convert my ideation STT as highly prompt-engineered coding instructions into @cursor_ai - My spoken thoughts are augmented with a very different type of Intelligence than mine. This will be so popular.

✨ Just released ✨ Faster and more reliable Offline Language Models Llama 3.2 running at 55 tokens/second on a M2 Pro

@sama @sama what it we consider RAG to be a closed source problem like math and coding? After all, the correct answer to a given (search)query is a function of the (correctly) retrieved chunk IDs. Wrote this paper on it last wknd. Reason to RAG: tinyurl.com/rag-linkedin-s…

trying GPT-4.5 has been much more of a "feel the AGI" moment among high-taste testers than i expected!

@iruletheworldmo I think you’re right. RL is going to change the game. Imagine if you apply GPRO to RAG agents by framing retrieval as a closed sourced problem. The correct answer are just chunk IDs after all. I wrote on it last week. wonderwhyai.notion.site/Reason-to-RAG-…

agi is coming. superintelligence follows. people still don’t get it. they think ai is just a fancy autocomplete, that it’s limited to regurgitating data, that it doesn’t “truly understand.” they hear sam altman say agi is near and assume he’s overhyping. as if a man running the most advanced ai lab on the planet, a man who has to actually deliver, would set expectations this high if he wasn’t certain. this isn’t speculation. the people at the frontier are not debating whether we’ll get there. they are watching it happen. ilya, sam, demis—all of them are telling you outright. reasoning is emerging. reinforcement learning is scaling. the models are getting smarter in ways no one predicted. this year we solve intelligence. after that, it’s just a matter of compute. it’s easy to miss how fast this is happening. a few years ago, ai couldn’t write a decent paragraph. now it can generate working code, diagnose diseases, and argue complex legal cases. soon it won’t just be helping engineers write software. it will be designing entire systems on its own. ai will automate medicine, law, finance, and anything else that runs on rules and reasoning. the bottleneck won’t be the intelligence of the models. it will be how quickly humans can adapt to the new reality. many of you will have your lee sedol, move 37 moment this year. you will watch something you thought was impossible happen right in front of you. entire industries will fall overnight. reinforcement learning will crack tools like operator, and the knowledge economy will become fully automated. ai won’t just assist. it will replace. this is what ilya saw. project q, strawberry, reasoning models, and the simplicity and power of reinforcement learning.* this is why he ousted sam hastily. he knew the moment was closer than anyone expected. as always, he sees further ahead than the rest of us. we won’t see it coming. we have been boiling the frog for years. every time ai crosses another milestone, we shift the goalposts. it can’t play chess. it can’t write. it can’t code. it can’t reason. every step forward is dismissed until there’s nowhere left to move the goalposts. by the time people realize what has happened, it will have already come and gone. agi won’t be a single moment. it will be a creeping inevitability, something we only recognize in hindsight, long after the transition is complete. and yet, people are still arguing about whether ai “understands” or whether it’s just a stochastic parrot. this is what happens every time a paradigm shifts. the old guard can’t see what’s in front of them because they are still anchored to the past. agi is coming. superintelligence follows. the walls are gone. the only thing left is whether you’re ready. soon, we will taste it, anon.

Here's my talk with Jason Liu @jxnlco - creator of the open-source #Instructor library, a fantastic framework for structured prompting, data extraction and validation (and more). We spoke on the October AI Builders Meetup #Amsterdam. We are a community of AI Engineers that hosts monthly meetups in Amsterdam, Rotterdam and Berlin. youtu.be/G2H9jNRO6Nc #LLM #Instructor #GenAI #OpenAI #Llama #Anthropic #prompting #aiengineer

Did @OpenAI just take a swing at @cursor_ai? openai.com/index/introduc…

@ivanleomk @jxnlco @jxnlco that’s awesome! Thanks! Keen to test it out.

Want to give O1 a try over the weekend? Just use instructor's new JSON_O1 mode and you're good. Same old familiar api with a single loc to change

@jxnlco Amsterdam! You get to join the infamous AI Builders Meet-up l, and I’ll get on the list to get into the coolest restaurants and clubs. 🤘

I totally delayed going to europe to read / vacation... where should i go? Paris, London, Berlin? Assume I'll be in Nov/Dec.

@trussliz @elonmusk @X Free speech means nothing without the ability to say something meaningful. Freedom without constraints is just chaos. Free speech without sensible restrictions is only anarchy.

I am appalled by the attacks on free speech in Britain and Europe. We can't be truly free without free speech. Good for @elonmusk and @X for standing up to these bullies.

@helloiamleonie That’s why we use query understanding -> transformation -> routing

Stop saying vector search retrieves the *most relevant* items to your search query. It doesn’t. It retrieves the *most semantically similar* items. Similarity and relevancy are not the same.

Generating interactive web components on the fly with the latest #openai model. got-4o-2024-08-06 is trained to return not only valid json objects, but also adheres with 100% accuracy to the your schema definition. loom.com/share/6914b82c… Code by @hive_echo

@shashwath19 @jxnlco After the designated field is populated I check if the value exists in the citation. If all true than pass, if false than retry kicks in. In the retry call I define a custom class to capture and analyse the validation error (so it does not get buried in context)

@shashwath19 @jxnlco I define a field to capture a citation as the source of the designated value for the next field. Than I normalize the citation and the source text of the input document. Exact match with fuzzy match fall back is used to verify the existence of the citation.

Instructor is dead tho.

@jxnlco You’ll be fine mate

If anyone is worried about my future after OpenAI steamrolled the marketing vehicle for my consulting business, please retweet.

@geertwilderspvv Heb je van je hobby je beroep gemaakt, loop je nog de hele dag te janken. Maar goed, daar ben je beroepszeiker voor. @geertwilderspvv, je zouteloze praatjes zijn als hergekauwde spruitjes stamppot met waterige jus. Slappe hap in optima forma.

Ongelooflijk wat een slappe hap dat OM hier, pro-Palestijns geweld tegen onze agenten loont dus is Nederland. Waanzin!