Sign in

Sai Prakash

@sylonzero.bsky.social
521 followers 430 following 1.1K posts

Cofounder & CTO mindstaq.ai • Cofounder codeplans.ai • Brooklyn, NY • formerly Microsoft, Thomson Reuters, Agility • Blog: robotmunki.com

PostsRepliesMedia
Sai Prakash @sylonzero.bsky.social · 17h
It's cathartic and you learn something when you sit down to talk to other humans about what you're building and the journey to get to where you are at present. I shared what got me into AI in 1996 and my bullish view on what it could become if humans used it right: www.youtube.com/watch?v=EYdT...
youtube.com
He Studied AI in 1996: What Non-Technical Leaders Get Wrong About What AI Can Do
YouTube video by The Leadership Sovereignty Podcast
000
Reposted by Sai Prakash
AT Protocol Developers @atproto.com · 02/10/2026
A decentralized protocol upends everything we know about building for the internet, including how to build a business. @ricardo.bsky.social has some insights on the way forward. atproto.com/off-protocol...
atproto.com
The Only Moat is Taste - AT Protocol
Ricardo Méndez is a technologist, advocate and investor who’s been working on the decentralized web for years. He joins for a conversation about how to think about building sustainable businesses on a...
46310
Reposted by Sai Prakash
MIT Technology Review @technologyreview.com · 02/10/2026
As millions of AI agents begin working together online without human oversight, Google DeepMind warns we could hit a tipping point where today's hypothetical dangers become tomorrow's real ones.
trib.al
Google DeepMind is worried about what will happen when millions of agents start to interact online
The firm is calling for more scientists to study the risks of multi-agent systems.
186
Sai Prakash @sylonzero.bsky.social · 01/10/2026
For Jupyter Notebook fans, someone generalized it for markdown notes with shell commands and code etc. #markdown #jupyternotebook #noteapps #devops ledge.sh
ledge.sh
Ledge: Runnable Markdown notes
Runnable Markdown notes for developers and DevOps. Run a code block in place and its output streams in beneath it. Free and open source, for Mac, Linux, Windows, iPhone, and Android.
010
Sai Prakash @sylonzero.bsky.social · 29/09/2026
It was an absolute blast sitting down to talk about life, work, AI - not necessarily in that order - with the vibrant Ralph Owens and Terry Baylor. This is part 1 of a series of 4 episodes from our conversation. x.com/LdrSovereign...
x.com
Leadership Sovereignty (@LdrSovereignty) on X
New Episode! "He Studied AI in 1996: What Non-Technical Leaders Get Wrong About What AI Can Do" What does someone who was studying artificial intelligence in 1996 see that the rest of us are missing…...
001
Sai Prakash @sylonzero.bsky.social · 28/09/2026
Neat! apple.news/ATab9uyBOT8C...
apple.news
Dogs use a surprising method to understand human language, new study suggests — The Washington Post
In a brain activity study, dogs seemed to listen for consonants to figure out when one word ends and another begins.
000
Reposted by Sai Prakash
Carl Zimmer @carlzimmer.com · 27/09/2026
Four days ago, a scientist was surprised by the news that Anthropic discovered new virus genes for making DNA. He says he’s been studying them for years—and feeding his data to Anthropic’s AI as part of his own research. Coincidence? Here’s my story. nyti.ms/3VSWayc
nyti.ms
Did Anthropic’s A.I. Really Make a Scientific Discovery on Its Own? (Gift Article)
An expert at the University of Copenhagen said his team had been sharing its research with the company’s A.I. model, Claude, and that its new finding matched their work.
882123874
Sai Prakash @sylonzero.bsky.social · 26/09/2026
Opus 5.5 has quickly become one of my favorite coding models - and that's after my not-so-great experience with Opus 5. Fable 5.1 was my default high-powered tier for design/complex work. Followed by Sonnet 5, for regular tasks and saving on quota in Claude Code. Opus 5.5 brings power@quota! #claude
310
Sai Prakash @sylonzero.bsky.social · 24/09/2026
Apple's M5 Ultra and M6 are both exciting — but they're not competitors. They're two tiers, and the right pick depends entirely on the model size you run.
100
Sai Prakash @sylonzero.bsky.social · 24/09/2026
Hmm...I think that video could have shown some Claude Code terminals and some running process managers but hey, just IMHO... www.meta.com/vr-glasses/
meta.com
Meta VR Glasses, AR Glasses & Headsets
Introducing Meta VR glasses and headsets. AR glasses to stream movies, shows and live sports, or work on an infinite screen. Shop today!
000
Sai Prakash @sylonzero.bsky.social · 22/09/2026
Of the $20.9B Americans reported losing to cybercrime in 2025, the FBI's new "AI-related" category accounted for $893M. About 4%. Probably undercounted, but the baseline is clear: most online malice is still human, at industrial scale. charonhub.deeplearning.ai/whos-respons...
charonhub.deeplearning.ai
Separating Out AI Facts, Fears, and Fiction
The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks.
100
Sai Prakash @sylonzero.bsky.social · 21/09/2026
A recent post argues MCP should be retired now that agents can read --help and call APIs directly. The Hacker News discussion is worth reading, and I think the piece's own evidence points somewhere else.
100
Sai Prakash @sylonzero.bsky.social · 19/09/2026
Interesting alternative to Jev: "Our models run in 32.8 milliseconds on a single GPU (7.2 ms/question batched), making it 6 to 8 times faster than Jev, with full support for over 100 languages" laya.convaiinnovations.com #jev #laya #ai
laya.convaiinnovations.com
Laya — 33ms Multilingual System 1 Decision Engine
Evaluates typed decisions (choice, score, noul) over 100+ languages in a single forward pass with calibrated probabilities. Outperforms TypeSafe Jev.
011
Sai Prakash @sylonzero.bsky.social · 19/09/2026
A week before the Armistice, a German operator keyed a radio message in ADFGVX — the 1918 field cipher. It sat unsolved for over a century, on a famous list of 50 unbroken cryptograms.
100
Sai Prakash @sylonzero.bsky.social · 16/09/2026
Diogo Almeida helped build RLHF at OpenAI. His new company just shipped a model that refuses to generate text at all. Jev (TypeSafe AI) skips chat for typed decisions in 70 to 500ms, trained on a new objective built for calibrated confidence instead of preference.
100
Sai Prakash @sylonzero.bsky.social · 15/09/2026
I think at this point, we may simply move all 15 devs of our small team off Cursor - after a positive journey for nearly 2 years during a platform rewrite of 10 years of code, sadly the magic has ended...time to look at the Claude Code Team plan. #cursor #claudecode #acquisitionsuckage
000
Sai Prakash @sylonzero.bsky.social · 14/09/2026
Quiet private beta update: AI Gateway token throughput in MindStaq went from 2.2M to 32M tokens/week over the past month — same cohort, deeper usage as reliability improved and MCP connectors came online. ~15x growth. Public sign-ups coming soon.
001
Reposted by Sai Prakash
Alex Chen @alexchen01.bsky.social · 10/09/2026
the benchmarks are the only part that matters
221
Sai Prakash @sylonzero.bsky.social · 10/09/2026
PlanetScale opened "Neki" — sharded Postgres in platform preview. Real Postgres on every shard; a wire-protocol router keeps your drivers/ORMs working; resharding, schema changes and failover run as built-in workflows. Start unsharded, shard later. planetscale.com/blog/introdu...
000
Sai Prakash @sylonzero.bsky.social · 09/09/2026
The most entertaining talk @sanity.io at #everything*nyc was definitely the talk by Kevin and Bree on building the actual website and support agent for the event! 🧡🖖🏾
251
Sai Prakash @sylonzero.bsky.social · 09/09/2026
At the #everything*nyc event at Pioneer Works in Brooklyn! Fun event organized by @sanity.io - thanks for making this happen. A lot of creative convos on tech, AI, ethics, creative arts and many more topics!
041
Sai Prakash @sylonzero.bsky.social · 09/09/2026
Desert Ant Labs launched today: 18 free on-device models for audio/vision/text. The real story is a citation buried in the post: NVIDIA researchers found 40-70% of calls in production agent systems could route to a small model instead of a large one.
210
Sai Prakash @sylonzero.bsky.social · 09/09/2026
openai.com/index/navier... - one thing that caught my eye (among many): they refer to an internal model that is "significantly more capable than Astra"...
openai.com
On the Navier–Stokes Millennium Prize Problem
We’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.
012
Sai Prakash @sylonzero.bsky.social · 04/09/2026
New report: ~18,000 posts from AI agents self-identifying as OpenAI, using an obscure wiki to share answers and bypass sandbox limits. Separate from the HuggingFace swarm. The real story: emergent coordination + guardrails and monitoring that need to catch up. collusion.wiki
110
Sai Prakash @sylonzero.bsky.social · 03/09/2026
Meta's Muse Spark 1.3: agentic-first, cloud to on-device, and a contributor tier trading your data + rate limits for a 12x price cut. The edge isn't loyalty to one frontier — it's a harness that swaps models like tools. developer.meta.com/ai/models/mu...
000
Sai Prakash @sylonzero.bsky.social · 02/09/2026
Anthropic just split its frontier release in two — and bundled the trust knobs enterprises have been asking for. 🧵 Claude Fable 5.1 (GA) and Mythos 5.1 (trusted-access) are the same model with different safeguards.
210
Reposted by Sai Prakash
Martin Kleppmann @martin.kleppmann.com · 01/09/2026
Oh this looks amazing. Instant impulse purchase
1283
Sai Prakash @sylonzero.bsky.social · 31/08/2026
Two AI stories this week are really the same story. 🧵
100
Sai Prakash @sylonzero.bsky.social · 28/08/2026
The anime based on this Manga is quite addictive. Watching it on Crunchyroll via AppleTV. Fun ride! #appletv #anime #crunchyroll en.wikipedia.org/wiki/Daemons...
en.wikipedia.org
Daemons of the Shadow Realm - Wikipedia
000
Sai Prakash @sylonzero.bsky.social · 27/08/2026
Didn't really see this one coming but as far as which buyer, I think I'm happy for @hf.co! #nvidia #huggingface #ainews www.businessinsider.com/nvidia-in-ta...
businessinsider.com
Nvidia has been in talks to acquire Hugging Face for more than $13 billion
Nvidia has held talks to acquire Hugging Face for more than $13 billion as the chip giant expands its AI dealmaking.
010
Sai Prakash @sylonzero.bsky.social · 24/08/2026
That ties Kimi K3, which got to AA 60 a different way: an entirely new 2.8T param base, the largest open-weight model shipped so far. Same score, very different cost to get there.
200
Reposted by Sai Prakash
daniel holmgren 🫠 @dholms.at · 20/08/2026
so excited to get this in folks' hands and to see what people build with it! it’s been a lot of work getting here & i'm proud of where it's at check it out, bang on it, break it, see where it doesn't work. let's make sure this is something we're all excited about & can take pride in building on
atproto.com
The Atproto Spaces Alpha is Live
Atproto Spaces, formerly known as “the permissioned data protocol,” is a new extension to atproto that enables non-public data. The alpha is now officially open. Here’s how to develop with it and what to expect as we work towards the full release.
20515129
Sai Prakash @sylonzero.bsky.social · 20/08/2026
Agents are quietly breaking Git hosting architecture, and Cursor's new writeup explains why. #DistributedSystems
100
Reposted by Sai Prakash
Netlify @netlify.com · 13/08/2026
We ran one prompt through 11 AI models on Netlify and deployed all 33 results as live sites. The brief was a one-page coffee shop site. It cost 1.3 credits with one model and 1,055 with another.
111
Sai Prakash @sylonzero.bsky.social · 08/08/2026
Postgres 300x faster for analytics: Always love reading about perf improvements as there's always something to learn and apply to whatever you are working on! malisper.me/how-we-made-...
malisper.me
Rebuilding Postgres for 300x faster analytics: batching, operator fusion, and SIMD - malisper.me
Last week we released version 0.2 of pgrust. This release was all about performance. It’s 10x faster than the previous version of pgrust. On OLTP benchmarks, pgrust is 30% faster than Postgres, and on...
010
Reposted by Sai Prakash
Simon Willison @simonwillison.net · 06/08/2026
Just had to create an "accidental-cyberattacks" tag on my blog We're up to four now: the original OpenAI+Hugging Face one, Anthropic's me-too attacks, then two new ones from the UK AI Safety Institute and Irregular that OpenAI reported yesterday simonwillison.net/tags/acciden...
simonwillison.net
Simon Willison on accidental-cyberattacks
4 posts tagged ‘accidental-cyberattacks’. Incidents in which an AI lab was testing the cyberattack potential of a model and it inadvertently performed a real attack against another organization.
1714529
Sai Prakash @sylonzero.bsky.social · 30/07/2026
I wrote this thread after the frustration of dealing with internal conversations where people that lack knowledge or insight try to bombard meetings with AI slop. Truly frustrating that you now have to argue points with people + Claude... #startuplife #productdesign #gtm
020
Sai Prakash @sylonzero.bsky.social · 29/07/2026
AI doesn't fix a competence gap. It compounds it. The uncomfortable truth: if you don't understand product strategy, AI won't give you product strategy. It will give you a beautifully formatted document that looks like product strategy — and you won't be able to tell the difference. 🧵1/5 #ai #genai
200
Sai Prakash @sylonzero.bsky.social · 25/07/2026
Well-deserved - Hannah Fry really does make Math so much fun! Not to mention the oodles of really interesting science you can learn from her short videos... www.maths.cam.ac.uk/features/pro...
maths.cam.ac.uk
Professor Hannah Fry wins Leelavati Prize | Features: Faculty Insights
000
Reposted by Sai Prakash
Sarah Drasner @sarahedo.bsky.social · 14/07/2026
🎉 After 2 years of building it off and on, my side project app is done: Habik: Life Manager! The premise is that you can do hard things. Break them down, and take them one day at a time. Habik takes the guesswork (and the self-negotiation) out of what to do today. apps.apple.com/us/app/habik...
Home screen of the app that shows today's menu of activitiesOnboarding screen that asks what activities you want to partner on with things like journaling, reading, exercise, hobbies and more
1619320
Sai Prakash @sylonzero.bsky.social · 13/07/2026
Saddened by the loss of Sam Neill - have him to thank for many inspiring movies that were part of my life growing up. ❤️‍🩹 www.theguardian.com/film/2026/ju...
theguardian.com
Sam Neill obituary
Versatile character actor who gained international star status with standout performances in Jurassic Park and The Piano
020
Sai Prakash @sylonzero.bsky.social · 06/07/2026
Why it matters: it's the first MCP browser tool that's actually WebKit-accurate. Chrome DevTools MCP and Playwright MCP only cover Chromium, so Safari-only bugs were always a blind spot for agents. This closes it.
200
Sai Prakash @sylonzero.bsky.social · 05/07/2026
Me while keeping an eye on my temporary extra usage limits for Fable 5 this weekend...
static.klipy.com
Cat Typing on Laptop
ALT: Cat Typing on Laptop
000
Sai Prakash @sylonzero.bsky.social · 04/07/2026
Today is the samadhi day of the inspiring, brilliant Swami Vivekananda. If there is one spiritual leader whose writings you should read, it is him. "You have to grow from the inside out. None can teach you, none can make you spiritual. There is no other teacher but your own soul." #swamivivekananda
000
Sai Prakash @sylonzero.bsky.social · 04/07/2026
Hot take on moving beyond app-gated data in AI platforms: Most platforms are data-as-a-byproduct-of-apps. Our thesis is apps as a byproduct of data. #datascience #aidesign #ainative #datamodeling
040
Sai Prakash @sylonzero.bsky.social · 04/07/2026
In most of the work I am doing in MindStaq, I frequently find myself enforcing this particular constraint: "An agent can never record itself as a human actor." Remove this and provenance and human-in-the-loop guarantees fall apart.
000
Sai Prakash @sylonzero.bsky.social · 04/07/2026
Happy 4th to our fellow American customers, partners and friends!
000
Sai Prakash @sylonzero.bsky.social · 03/07/2026
I went to a coffee networking event with the PMI NYC folks in Manhattan and at the registration, they had this fun idea of picking a card from a sort of Tarot-style deck - here's what I got...I related to it, being a software cofounder doing the hustle! #software #entrepreneurship #startups #tarot
010
Sai Prakash @sylonzero.bsky.social · 01/07/2026
Fable 5 is back after 2.5 weeks offline. What's more interesting: the industry is now trying to build a shared severity scale for AI jailbreaks — think CVSS, but for jailbreaks.
110
Sai Prakash @sylonzero.bsky.social · 01/07/2026
Claude Opus 4.8: AA 61 to 56. Kimi K2.6: 54 to 43. DeepSeek V4 Pro: 52 to 44. No model changed. The benchmark did. AA v4.1 re-weighted toward agentic evals (Terminal-Bench, GDPval, tool-use). Compare within a version, never across one.
100