a.t. yılmaz (@_streamerd) header
a.t. yılmaz (@_streamerd) profile picture

intentions precede impressions

Down to Earth
Joined November 2009
Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)Photo by a.t. yılmaz (@_streamerd)
a.t. yılmaz retweeted
finished an independent full verification of Levent Alpoge's Lean 4.33.1 proof of the Riemann Hypothesis and Dirichlet GRH, and it passes both Lean and an independent Rust proof checker (nanoda) with zero errors. what actually got verified: > compiled all 175,187 reachable files (17.37M lines) from the raw alpo.ge 14-part archive using unmodified stock Lean 4.33.1 in ~2 hours on a single rented 192-core box ($12.53 total spend) > pinned all five headline theorems directly against unmodified upstream Mathlib (RiemannHypothesis + four Dirichlet L-function GRH theorems), zero sorries or custom axioms, strictly propext, Classical.choice, and Quot.sound > exported all five theorems' full proof closure (61.57 GB / 1.08 billion lines) and streamed it into nanoda, an independent Rust kernel with zero shared code from Lean: 1,484,729 declarations checked, 0 errors, same three standard axioms only how it ran fast without touching the compiler: > bypassed lake's ~943 GB .setup.json overhead with a direct stock-Lean DAG runner using exact lakefile.toml flags > strace showed Mathlib sitting 9th in LEAN_PATH caused ~300k failed statx calls per file; merging packages into 1 dir per NUMA /dev/shm shard cut kernel time from 29% to ~10% and jumped throughput from 14 to 25 modules/sec caveat: this settles the five formalized statements (RH + Dirichlet L-functions, not general Dedekind/automorphic GRH) at the machine-kernel level under standard axioms. humans still have to read and understand what the proof is doing. i've linked receipts, hashes, and reproducible runner in the reply.
18
65
12
667
61,791
a.t. yılmaz retweeted
Andrej Karpathy: “A 1B model can be much more effective than most current LLMs. 90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.” here is his 5 core thesis about Small LLM’s: 1. the intelligence was never the knowledge. Strip the memorized internet, keep the reasoning. What’s left is the cognitive core, and it’s small. 2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain. 3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count. 4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone. 5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you. in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability. TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex. It’s the thesis Karpathy’s been repeating all year, made real. study more on this below ↓
Modest intelligence is arriving faster than most people realize. Today, we're open-sourcing TwIL-8B. Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks: • 93% of frontier performance on coding • 92% on instruction following • 90% on contest mathematics All from a model small enough to run on your own hardware. But the benchmarks aren't the most important part of this announcement. For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it. I believe that equation is changing. The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task. If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model? As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own. That's the future we're building at @thewebAI TwIL-8B is available today. Open weights. And this is just the first of several announcements we're making today. HF Link: huggingface.co/webAI-Officia…
205
660
136
7,002
1,290,985
I believe I can try
28
a.t. yılmaz retweeted
13 Ekim. Bugün babam Aykut Oray’ın doğum günü. “Vatandaşa cart curt yok” sözüyle hafızalara kazınan o fiyakalı beyefendi, namıdiğer “Katil Yavuz”, yaşasaydı 84 yaşında olacaktı. Babam tiyatroya yıllarını vermiş, çok yönlü, meraklı ve yeniliğe açık bir oyuncuydu. İstanbul Üniversitesi Ön Asya Dilleri ve Kültürleri Bölümü’nden mezun olmuş, Hititoloji üzerine doktora çalışmaları yapmıştı. Çivi yazısı okuyup çevirebilen bir akademisyendi. Tarih, coğrafya ve antropoloji bilgisi muazzamdı. Türkiye’nin neresinde ne yapılacağını, ne yenileceğini ve nereye gidileceğini çok iyi bilirdi; onun muhabbetinde hem güler hem bir şeyler öğrenirdiniz. Mustafa Kemal Atatürk’e ve Cumhuriyet’in ilkelerine bağlı, ülkenin bölünmez bütünlüğünü savunan bir sosyal demokrattı. Doğruluktan, dürüstlükten ve hukuktan yana olur, kimden gelirse gelsin, haksızlığa ve hukuksuzluğa karşı çıkardı. Cart curt yaptırmazdı. Fanatik bir Beşiktaşlıydı; yürüyen Beşiktaş’tı. Torunlarının nasıl bir dedeleri olduğunu, hangi değerleri savunduğunu ve nasıl yaşadığını bilmeleri için elimizden geleni yapıyoruz. Onun evlatları olmaktan ve insanların kendisini hâlâ sevgiyle, saygıyla ve güzel anılarla hatırlamasından gurur, onur ve şeref duyuyoruz. Mekânı cennet, ruhu şâd olsun.
566
442
38
8,965
154,222
We are entering an era in which people are enabled to make things for their own use rather than motivated to sell to others.
28
software orchestrator
35
I stopped saying IRL and real-time. Both recently started introducing new kinda oxymorons.
37
got my first pod at consauth.com looking cool, uh? #expressionofdemand
36
Art is the art of creation
30
word of this year shall be: sovereignty
1
44