Maciej Obarski @datageek_ai

applied AI / CS / tech insights 🎯 target shooter, sci-fi geek 🚀 techno-optimist github.com/mobarski Warsaw, Poland Joined December 2014

Tweets

3K
Followers

898
Following

1K
Likes

15K

🔥 Matt Dancho (Business Science) 🔥 @mdancho84

2 days ago

OpenAI, Google, and Anthropic just published guides on: • Prompt engineering • Building agents • AI in business • 601 AI use cases 9 of the best guides you can't miss:

12 59 349 27K 596

Download Image

Aran Komatsuzaki @arankomatsuzaki

3 days ago

Meta introduces Set Block Decoding (SBD), a new inference accelerator for LLMs SBD samples multiple future tokens in parallel, cuts forward passes by 3–5x, needs no arch changes, stays KV-cache compatible, and matches NTP training performance.

4 23 162 18K 121

Download Image

jason liu @jxnlco

3 days ago

If your boss is asking you to prototype an agent, I think this is the best way to get started. jxnl.co/writing/2025/0…

13 48 491 32K 782

Casper Hansen @casper_hansen_

a week ago

xAI may be one of the single biggest contributors to open-source inference just by serving everything with SGLang

12 34 554 197K 179

Download Image

OpenRouter @OpenRouterAI

2 weeks ago

Fourth model launch of the day 🔥 - introducing Hermes 4, from @NousResearch Hermes 4 is trained for steerability and lower refusal rates, topping RefusalBench and beating Grok 4

10 37 380 39K 61

Download Image

Jackson Atkins @JacksonAtkinsX

2 weeks ago

NVIDIA research just made LLMs 53x faster. 🤯 Imagine slashing your AI inference budget by 98%. This breakthrough doesn't require training a new model from scratch; it upgrades your existing ones for hyper-speed while matching or beating SOTA accuracy. Here's how it works:…

91 697 4K 430K 4K

Download Image

Cline @cline

2 weeks ago

Our engineer @arafatkatze explains why simpler approaches consistently outperform architectural complexity: cline.bot/blog/3-seducti…

3 5 36 6K 28

Rohan Paul @rohanpaul_ai

2 weeks ago

Another paper claiming really BIG result. The First method to achieve 99.9% on AIME 2025 with open-source models! 🤯 DeepConf uses a model’s own token confidence to keep only its strongest reasoning, with GPT-OSS-120B while cutting tokens by up to 84.7% compared to standard…

20 157 831 69K 886

Download Image

Joshua Burgin @joshuaseattle

2 weeks ago

David Anderson is right on. This is about more than avoiding burnout. It’s the core mentality difference between leaders & companies that scale and those that don’t. At Amazon & AWS we talked about “big rocks” - the few critical priorities you protect time and focus to deliver.

1 11 235 49K 542

Download Image

Adrian | The Web Scraping Guy @adrian_horning_

2 weeks ago

The legality of scraping twitter in 2025: Anything in front of the login (public), is legal and fair game. Problem is, the real juicy stuff (latest tweets, search) are behind the login :( Let me give you the rundown of some recent history and what your options are:

8 6 116 20K 185

Rohan Paul @rohanpaul_ai

2 weeks ago

MASSIVE claim in this paper. - 10–12.5x memory savings vs. FP16 - Near-zero accuracy loss - Beats state-of-the-art KV quantization🔥 XQUANT, a way to break the memory wall for LLM inference via KV cache, by storing inputs and recomputing attention on demand. Key Proposals:…

11 72 404 31K 316

Download Image

Maxime Rivest 🧙‍♂️🦙🐧 @MaximeRivest

3 weeks ago

Today, I pruned 87.24% of Qwen 30B for a sentiment classification task while keeping 100% of its accuracy. This means we get to use big models on gpus with not that much RAM (potentially running models that would normally require an H100 on 3090 type gpus)! Imagine pruning…

78 154 2K 184K 2K

Download Image

William J.B. Mattingly @wjb_mattingly

a month ago

I spent the last 24 hours finetuning Dots.OCR with different datasets. Here are some of the things I learned...

13 47 557 63K 692

Download Image

Maxime Labonne @maximelabonne

4 weeks ago

Liquid just released two 450M and 1.6B param VLMs! They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion. Available today on @huggingface!

17 83 642 103K 351

Download Image

Jack Morris @jxmnop

4 weeks ago

OpenAI hasn’t open-sourced a base model since GPT-2 in 2019. they recently released GPT-OSS, which is reasoning-only... or is it? turns out that underneath the surface, there is still a strong base model. so we extracted it. introducing gpt-oss-20b-base 🧵

159 470 6K 919K 4K

Download Image

👋 Jan @jandotai

4 weeks ago

Introducing Jan-v1: 4B model for web search, an open-source alternative to Perplexity Pro. In our evals, Jan v1 delivers 91% SimpleQA accuracy, slightly outperforming Perplexity Pro while running fully locally. Use cases: - Web search - Deep Research Built on the new version…

102 446 4K 675K 3K

Download Video

All Hands AI @allhands_ai

4 weeks ago

We evaluated GPT-5 in OpenHands and it's the new number one coding agent model for us! Using exactly the same tools and harness it's 1.4 points better than Claude Sonnet 4 at 60% of the price. Full results here: docs.google.com/spreadsheets/d…

14 31 247 31K 73

Download Image

Binyuan Hui @huybery

a month ago

We'll continuously enhance the qwen code (cli tool) based on your feedback and even release improved qwen-coder (model)! Our goal is to match Claude Code's performance while remaining fully open-source!

Qwen @Alibaba_Qwen

a month ago

190 636 5K 913K 4K

Download Image

148 164 3K 174K 613

Oscar May • Quntem @TheOtherSeru

a month ago

i think about this clip from @theo a lot

31 79 983 70K 592

Download Video

AI Tomorrow @aitomorroww

a month ago

@ryolu_ don't replace human connection don't automate away meaning don't optimize for engagement over truth don't create dependency without understanding don't build surveillance disguised as assistance don't monetize human insecurity don't accelerate without wisdom don't…