Turned my ancient portfolio (basic HTML/CSS) into a modern webpage with Next.js + TypeScript + Shadcn + GSAP.
Old junk: https://t.co/xjnMoKOprt
New site: https://t.co/shgXnsc51p
all with AI (love @claudeai Code). 🫡
for anyone thinking “AI can now make models smarter”: read the post. Astra found what made a 0.8B model loop and removed it, so it used math skill it already had. on Gemma 270M (no loops) it couldn’t improve math. removes obstacles, doesn’t add abilities. cool, not AGI
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq
@rikiyatakehi@a1zhang@lateinteraction baseline had 93% of answers hitting the token limit, so 7.9% MATH-500 is mostly a looping model, not a model bad at it. did u compare against Qwen’s recommended sampling or a repetition penalty? if decoding fixes close most of the gap, edit is much smaller result than headline
project-1: https://t.co/66CwZXnxTA
I built a Chrome-Dino game that's really a tiny game env over WebSocket: a bot can play. the model plans ahead instead of reacting. It now plays at the same speed a human does, makes ~one call per obstacle, and out-runs the game indefinitely.
tokenmaxxing because of pure curiosity.
A glimpse of how I spent 10.08B Claude Code tokens in just 52 days:
• AI trading agents
• Hacking Alexa & home automation
• BLE reverse engineering a smart band
• Autonomous drones
• MCP servers
More on each of these projects soon..
We've grown 15x since our last funding round.
Today, our agents support 3,200 providers across 1,500+ sites in 27 states. 1 in 50 Americans can access care through them.
We're building healthcare software that takes action, giving existing teams more capacity to care.
We're just getting started.
https://t.co/7rWOV5snqG
We've raised $22.5M in total Series A funding for Confido Health, including a new $12.5M extension led by @z47_vc.
Medicine has added decades to our lives, yet patients still lose weeks to paperwork.
We're building AI agents to give healthcare teams more capacity to care.
I'm in! 🔥 Joining the @lyzr__ai Agentathon - April 25, Bengaluru.
500 builders. 5-hour sprint. Building the next era of agentic systems.
Apply for your spot 👉 https://t.co/XV3UQJn0Mg
#LyzrAgentathon#AgenticAI
@GithubProjects Yup! Here’s mine:
> https://t.co/yv4D0hApRr
Been tinkering with AI, full-stack, automation and a bunch of experimental stuff w ~180 projects so far and still cooking 😄
The only way to celebrate @_groww’s listing day.
I got bored, reverse-engineered my smart lights, and hooked them to the Groww Cloud API.
Here's the video of my entire room turning GREEN as $GROWW's price went up today.
It's not 1 light. It's 3.
#Groww#GrowwIPO#StockMarket
Took this idea further 😎 reverse-engineered @Philips_lights, mapped price delta to hue/brightness, and plugged into @_groww trading api. Three bulbs in sync; #GROWW pop turned them GREEN. Open source Repo: https://t.co/zU7vLv3FJF
This was all a single-night hack.
Massive shoutout to the @VamsiK76294 (#Groww) team for giving me free access to their Cloud API to build this.
This is officially the most over-engineered "HODL" indicator ever.
GitHub Repo: https://t.co/zU7vLv37U7
#Trading#Tech#WizHack
The "Hack": These Wiz lights have zero security. They just accept raw packets on UDP port 38899.
A Python script hits the @_groww API, does the math, and blasts UDP commands to all 3 lights in real-time.
The best part? The lights also flash if the price hits my stop-loss. 😎