Cofounder & CTO https://xitter.cf/t.co/SsAv0DJQS3 • Cofounder https://xitter.cf/t.co/0In4mwY8NF • Brooklyn, NY • formerly Microsoft, Thomson Reuters, Agility • Blog: https://xitter.cf/t.co/KnZX8DMqlp
New York, NY
Joined March 2026
- Tweets669
- Following223
- Followers34
- Likes970
Sai Prakash retweeted
A 40B coding model running more than 125 tokens a second on a MacBook, at short contexts. Open weights, Apache 2.0.
@recursive_si had their own AI research system help build it, from scratch, in under four months.
I've wanted to do all my coding locally for years.
This gets close.
🤝 Paid partnership
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
Sai Prakash retweeted
We open sourced a *really fast* HTML artifact builder. It's powered by libfx, making the agent durable.
Building the low-level plumbing for v0 to survive server crashes, flaky LLMs, function limits was a huge pain.
Now it's easy, free, and OSS: fx0.vercel.dev.
PS: I built this floating container in a few minutes… it's delightful, and it's just HTML!
We built a 'mini v0' on libfx to showcase its speed and durability:
• Design or prototype anything in seconds
• 200+ models from Vercel AI Gateway
• Outputs are pure HTML artifacts
• Fork it & make it your own
Try it free ↓ fx0.vercel.dev
Sai Prakash retweeted
"I want the executive to have a dashboard light up" is "not a motivation in itself."
Sai Prakash, CTO and co-founder of MindStaq, on the hardest lesson in adoption.
leadershipsovereignty.com/ep…
Sai Prakash retweeted
Hello Software Engineers, I lead TypeSafe AI Model Capabilities team. We are hiring. Here are a few qualities we are looking for:
• strong plus for people who resonate with the title "Experience Engineers": mobile engineers, full stack engineers, or even (virtual) reality engineers.
• strong coder (you need to be able to manually write code)
• product oriented thinker
I promise are fun! I trust that you will figure out how to apply.
Sai Prakash retweeted
our CEO @CompleteSkeptic tells the tale of Dr. Einstein and Mr. Bean
> if AI responds like Einstein 95% of the time and Mr. Bean 5% of the time, how are you going to automate anything with it?
@mattturck talks us through a wild first month of Jev
Crazy launch! Copycats! Massive round! On the one-month anniversary of the launch of Jev, we unpack it all with @CompleteSkeptic, CEO of @typesafeai: what Jev is, how it fixes what's broken in AI, and why the fast-followers are missing the hard part.
00:00 “95% Einstein, 5% Mr. Bean”
1:09 Why intelligence per dollar matters
3:25 Why TypeSafe built in stealth for two years
5:50 “We could have had AGI last year”
6:48 Why the product carries the marketing
9:33 What is Jev? Machine-native AI, built for software
10:53 System One vs. System Two
12:26 From simple decisions to composable AI
15:21 Automating email: giant prompts vs. smart if-statements
20:03 Probabilities, thresholds, and developer control
23:43 Why Jev is faster and cheaper than chat models
25:39 “Classification is AI-complete”
27:16 RLHF: how OpenAI trained models to please humans
31:28 Sycophancy, refusals, and the hidden costs of RLHF
34:22 RLVR: why benchmark gains don’t mean reliable automation
37:51 RLCD: optimizing for machine-native intelligence
39:54 Why TypeSafe builds its training data in-house
41:49 The Bitterest Lesson: define the task before scaling
44:39 Building Jev from open-weight models
45:56 Is Jev just a classifier? Why that’s a compliment
48:35 “Build prod, not God”
50:11 Embedding AI in software, beyond coding assistants
53:05 Jev’s copycats: matching the interface vs. the intelligence
55:33 Why benchmarks mislead the users they should help
58:37 Open weights vs. APIs: how TypeSafe weighs the tradeoff
1:00:04 Jevons paradox: a new market for machine intelligence
1:02:50 Why Diogo is skeptical of recursive self-improvement
1:04:47 Can practical automation avert an AI winter?
Sai Prakash retweeted
"Enhance every person so that they're the hero of their role. But don't try to replace the role itself."
Sai Prakash, CTO and co-founder of MindStaq, on the line AI should not cross.
leadershipsovereignty.com/ep…
Sai Prakash retweeted
Tune in to our town hall in Discord at 1:14pm
discord.gg/typesafe
what's this?
Jev is evolving!
tune in to our town hall with @CompleteSkeptic and @allietheicon at 1:14 PM to learn more
Sai Prakash retweeted
"How do you get adoption? Are you going to use a carrot or a stick?"
Terry Baylor on the onboarding that taught him how not to roll out a tool.
leadershipsovereignty.com/ep…
Sai Prakash retweeted
UltraBERT > Ettin BERT > ModernBERT > BERT, RoBERTa
UltraBERT is the latest BERT model outperforming both Ettin BERT and ModernBERT models
UltraBERT is the first encoder to achieve state of the art on both classification and retrieval tasks, reaching 91.0 on GLUE and 47.7 on BEIR.
Key Features
- Uses two masking ratio (5% helps classification tasks and 50% helps retrieval tasks)
- Use two separate LM heads (one for 5% masking ratio and another for 50% masking ratio)
- Has a context length of 32K tokens (4 times the context length of ModernBERT)
- Available in two sizes: 150M and 380M.
- Trained over 2 trillion tokens
- Uses Knowledge distillation from 1.3B teacher to train 380M and 150M models
- Open Source and weights are available on Hugging Face
Paper - arxiv.org/abs/2610.15028
Model weights (UltraBERT 150M) - huggingface.co/dmis-lab/Ultr…
Model weights (UltraBERT 380M) - huggingface.co/dmis-lab/Ultr…
#UltraBERT #deeplearning #datascience #ai #NLProc
Sai Prakash retweeted
"The hero of the story is not your app, it has to be the user."
Sai Prakash, CTO and co-founder of MindStaq, on why project tools get ignored.
leadershipsovereignty.com/ep…
Sai Prakash retweeted
New Episode! "Hero of Their Role: How to Get Team Buy-In on Project Management Without Chasing Updates"
Why won't your team keep the project tool updated, and what would make them want to?
Most pro…
Player links & show notes: leadershipsovereignty.com/e/…
Sai Prakash retweeted
A less discussed and potentially game-changing aspect of WebMCP: helpful a11y.
Assistive tech struggles with the same interfaces agents do. Dense menus, unlabeled buttons, etc.
When a site has clear tools, we can expose exactly the functionality we need.
Sai Prakash retweeted
You can literally type a sentence and watch a Transformer process it.
This interactive explainer runs a GPT-style model in the browser and lets you follow the whole path:
tokenization → embeddings → Q/K/V → attention → MLP → logits → next-token probabilities
You can inspect attention maps, intermediate values, and change temperature, top-k and top-p to see how generation changes.
If Transformers still feel like boxes connected by arrows, this is worth opening.
poloclub.github.io/transform…
Sai Prakash retweeted
Sai Prakash grew up on Isaac Asimov, and it shaped how he builds.
"People were always central to that thesis." The machines existed to make human life better.
Who is your technology built to serve?
leadershipsovereignty.com/ep…
Sai Prakash retweeted
Choose d1-3B for high-quality text and vision decisions, or explore our experimental d1-omni-600M when footprint matters.
Download the weights, fine-tune, and deploy locally.
> Blog: liquid.ai/blog/d1-open
> d1-3B: huggingface.co/LiquidAI/d1-3…
> d1-omni-600M: huggingface.co/LiquidAI/d1-o…
> Docs: docs.liquid.ai/lfm/models/de…
Sai Prakash retweeted
massCode - 免费开源的开发者工作台。
集代码片段、笔记、HTTP 请求、计算器、绘图、开发小工具于一体的本地优先应用。
你的工作空间以纯 Markdown 文件保存在本地。
👉 github.com/massCodeIO/massCo…
Sai Prakash retweeted
Cloudflare's Clef beat Jev in just two weeks
How: "frozen" Qwen does one prefill pass, a tiny schema head scores every answer option in parallel, with no text generated at all.
That's why it's 4x faster than Jev at 2x the accuracy on some benchmarks.
And it's open source!
Link to HF:
huggingface.co/Cloudflare/cl…
Sai Prakash retweeted
Run Qwen3.8 27B TensorFold on a DGX Spark
Default settings:
- 256k context by default
- Can be extended to 1M with YaRn
- 2.5K KV cache
- 8 concurrent session
Performance:
- 58.3 tok/s on prose, single stream
- 108.5 tok/s on prose, 2 concurrent streams
- 220 tok/s on prose, 8 concurrent streams
That puts single stream 2x faster with TensorFold!
mia-ai.net/models/Qwen3.8-27…
Sai Prakash retweeted
Are LLMs just glorified text predictors? 🤔
They get better over time, but they don’t reason.
What do you think?
#AI