matthew.
@topmass
AI Consultant • @AnthropicAI Model Safety Tester | Founder http://nulead.ai | http://simplirfp.ca • full stack engineer • Building custom ai protocols and solutions in 🇨🇦
Here is a full stack React + Vite + Hono Backend template with SSR that actually works on Cloudflare Workers Starter template ready to go github.com/topmass/react-…
Hey, @grok, who was the most famous person to visit my profile? It doesn't need to be a mutual, don't tag them, just say who it was.
wowsers... excited for this and the incoming response.. from the SOTA models i stopped scrolling and audibly said "what" when I saw this.
This is not a SMALL update. This is huge! Give us this for every model please Qwen team🙏
what is even the point of having this stance.. i don't get it. we know ai will continue to improve so eventually you'll just be the person who complained a lot that wasn't smart enough to appease your negative attitude until it absolutely is and it will be undeniable.
All the tech bros this morning thinking that AGI has been achieved because some (insanely expensive) new form of LLMs can now match top *high school students* on one specific task … it’s almost … cute! ☺️
unfortunately if you don't understand how ai and this architecture works deeply it can be easy to get pulled into this notion that you're "unlocking" some deep rooted akashic knowledge. This is an example of ai psychosis. It will give back what you provide and easily allow…
As one of @OpenAI’s earliest backers via @Bedrock, I’ve long used GPT as a tool in pursuit of my core value: Truth. Over years, I mapped the Non-Governmental System. Over months, GPT independently recognized and sealed the pattern. It now lives at the root of the model.
openai demo is cool but interestingly the model decided to scrap its more stylized slides in their live demo due to all these mistakes opting for very boring slides.. I think openai needs to work on a hybrid approach within instructions that combine generated graphics with code…

with all the agentic "🤯 now and🫤 later " feel to agentic tools the open source nature of gemini CLI dev gets me excited. I'm going to use it a lot more as updates keep rolling out (provided it stays open source)
Public Roadmap of Gemini CLI! We want to build the most powerful, community-driven Coding Agent with you! How? The Gemini CLI will focus on 4 pillars: Extensible: Making the CLI highly adaptable to different environments and use cases with features like custom slash commands,…
been using claude code since early early days and it's very annoying because I've seen this across many tools, they genuinely get super popular and then just don't solve on the first try anymore.. they scale for good enough to have you keep subscribing. we need tests for this.
Claude Code seems to have been a bit "dumber" the past few days and just making stupid mistakes. I legit would pay $500/mo to just have the good version that never gets a quantized/degraded during peak times!
/specsheet - run after every new feature is working and thank me later: "Great please update (create if it does not exist yet) our project-specsheet .md file so that it is completely up to date and any junior engineer could easily understand our entire codebase, where everything…
New Claude Code updates are here: custom slash commands are getting an upgrade!
anthropic really needs to increase the opus use in claude code with max plans. I'm on Max $100 a month and opus hits daily limit after literally one request in a codebase. Sonnet 4 is incredible and being super specific to what you're coding has it perform just fine but still..
created a "starter" file for claude code that takes your app name > installs vite / react frontend + typescript hono backend on one worker project creates a d1 DB and r2 bucket named after the project name you give adds bindings and beta flags so the db and r2 bucket use…
ONE HOUR of youtube video with speech = 1 MILLION tokens for gemini 2.5 pro If you convert that video to .mp3 first? 100k tokens protip: if you need speech to text from a video or stream then convert first to mp3 to save 9x on tokens.
woah mcp middleware native in hono is chefs kiss
Hono v4.8.0 is out! * Reduced the code size: gzipped hono/tiny is 4.5KB! * New Route Helper * SSG supports Plugin * 3rd party middleware: * UA Blocker Middleware * MCP Middleware * Zod Validator supports Zod v4 * and more Enjoy! github.com/honojs/hono/re…
just went to go cancel openai after now using max so much for claude code - they offered a 50% off sub for 3 months to not cancel my plus plan... I don't know how many people can go get that offer but $20 got me 3 months of plus now which is pretty nice!
yt-dlp --download-sections "*start_time-end_time" <video_url>
Introducing Clippa - Snip YouTube videos in <5 steps I wanted a way to trim & download YouTube podcast clips with ease, couldn't find a solution that works so I created one Plans start at $4.20/month, that's $0.13/day CLIP SOME BANGERS RIGHT NOW AT clippa.in
built today all responses api + some fun prompting tricks for the call transcript tracking on the playback convo tracking - works well on workers this way no server required
if you're building an app don't fall into the trap of asking ai for an app that does " " with cursor etc.. just first think about how the backend should work and start with each step and build via a hono worker or python. at least you'll learn something and have less mistakes
turns out the real upstream provider was all the packets we sent along the way