pymike00 (@pymike00) header
pymike00 (@pymike00) profile picture

Retired software craftsman. Now sailing the seas of the AI revolution.

Joined February 2016
Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)Photo by pymike00 (@pymike00)
Mamma, butta la pasta Now approved in Italy 🇮🇹 Rollout begins soon
732
1,800
828
13,390
1,007,505
pymike00 retweeted
Replying to @loktar00
I mean look at THIS. 20 t/s on my M4 Pro 48GB via Strata. It also did better than 27B optimized Pi on a scheduler benchmark with Astra as a judge.
1
1
23
pymike00 retweeted
We shrank 1 GB of Wikipedia to 105.39 MB, losslessly. Introducing tufazip: the best entry on the Large Text Compression Benchmark using a neural network as its sole predictor. Everything above it mixes a transformer with hand-built models. Full post 👇 tufalabs.ai/research/tufazip
449
873
486
12,635
2,487,864
pymike00 retweeted
a beautiful mid sized model is coming up it's not the Qwen you haven't heard from me
getting back to shipping, november will be a good month for open models 🫶🏻
32
6
1
376
14,857
Inkling 2 Small + Local Code + M5 Ultra = 🇺🇸 Available now in Local Code 26.1118.1 (self-distributed). DSpark support is coming. If you have strict requirement for all-American model usage / deployment, Inkling 2 Small is a great choice. DeepSeek 4.1 Flash level of intelligence, good token-efficiency, great visual capability.
2
3
1
11
1,507
pymike00 retweeted
show this video to anybody that is an old school programmer that hates AI this is one of the best videos I’ve seen all month.
1,077
6,264
1,250
45,710
3,817,872
Ragazzi sto assistendo al litigio più divertente dell’anno. Michele Boldrin è tutto il pomeriggio che litiga con Grok. Penso rimarrà negli annali di X. Prof @ricpuglisi non se lo perda.
No. La legge adotta la definizione IHRA (con esempi) solo «ai fini delle attività e delle disposizioni» di questa norma: strategia, coordinatore, formazione, banca dati. Non modifica l’art. 604-bis c.p., non crea reati né sanzioni nuove. Il codice penale resta identico. L’art. 1 tutela esplicitamente la libertà di critica politica. Quegli esempi restano opinioni politiche, non automaticamente illeciti.
75
54
3
493
15,795
The fact that this guy was kicked out of his company is the epitome of the current state of society
Jensen Huang wasn't predicting the future, he was literally creating it. (2009)
69
pymike00 retweeted
How much can you improve a small model in 18 days on 144 GPUs? Meet DaisyRose-27B, RadixArk's first open-weight model! It is trained with RL on Qwen3.8-27B, with 1,200 curated long-horizon agent tasks for 18 days on 144 H200s. Powered by Miles and our own training backend, we stretched agent sessions up to 1M tokens and raised the Terminal-Bench 4.1 score from 5.6% to 12.0%. See how we pull it off ⬇️
60
116
62
815
560,027
If you’re reading this, you’re in the top ~0.005% on earth for understanding how many legacy systems are about to get routed around.
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included). Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs. Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines. This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!! Enjoy: science.caseyhandmer.com/phy…
34
102
2
2,796
136,568
Local AI is wild. Hardware is REALLY being pushed and so many optimizations and workarounds are being put in place! I am sure we are gonna see so much improvement over the following months! AND soon we are getting Qwen 4 !
2
2
94
My project Slotstream hit 500 stars on GitHub today! Super excited! I've been talking with a lot of people and getting feedback and insights. We really are in the Homebrew Computer Club days of local AI
1
1
6
120
pymike00 retweeted
Qwen4-35B👀 Wait how the fck did this go under the radar !! @shuai_bai_ who leads Qwen-VL & Visual Intelligence replied this !! what else is this supposed to mean other than possibility of us getting a Qwen4-35B !! or some other smaller models ? Make it happen @QwenDevs !!
Replying to @IdolHomeless
Who said we forgot about the smaller ones? 👀
28
9
5
427
31,573
Introducing Qwen3.8-27B-pi-adapted 🤗 We adapted Qwen's model specifically for the pi coding agent using reinforcement learning and a LoRA adapter. The results, measured as pass@1: 🚀 SWE-bench Pro (Hard 51): 29.4% → 62.7% 💻 Terminal-Bench 2.1: 39.5% → 55.8% ⚡ 55–62% fewer tool calls per task This is the power of open models: you can adapt them to YOUR tools and YOUR use cases. Let’s build! 💛 Open weights: huggingface.co/harness-hub/Q… Methodology: huggingface.co/blog/harness-…
304
643
173
5,991
452,307
pymike00 retweeted
It is not model-maxxing. @localcodeapp respects your available RAM and you can do other things while keeping Local Code working. Using a smarter model like DeepSeek 4.1 Flash 1017 is great because you don't need to think about "how the model would do the work if it were me", you just need to think what outcome you want from the model and ask the open question.
DeepSeek v4.1 Flash on Mac with 32GB? Yes you can using SSD streaming and Local Code! Here with 64GB settings. getlocalcode.app
3
1
17
2,360
Testing deepseek 4 on a 48GB mac, it works slow, but it works! @localcodeapp is awesome!
DeepSeek v4.1 Flash on Mac with 32GB? Yes you can using SSD streaming and Local Code! Here with 64GB settings. getlocalcode.app
2
98
pymike00 retweeted
Claude reverse engineered my 32-year-old Jeep. Then it started controlling it. Edited with Fable 5.2
631
1,779
636
19,396
1,801,267
pymike00 retweeted
Geniale rappresentazione del perché l’industria italiana stia miseramente fallendo.
245
616
204
7,397
947,984