Oleg Ciubotaru
Tech commentator who engages with the scale of resource extraction (e.g., copper demand vs recycling limits) as it relates to electrification, data centers, and AI.
AI Services Sacrifice Five‑nine Uptime for Double Capacity
Great Stanford CS153 course... Amin Vahdat who runs Google's AI infrastructure and the TPU program behind Gemini. His point on reliability: "Five nines (99.999%) = 30 sec of downtime/year - but it costs you 2N redundancy, so half your power sits idle. Three nines (99.9%) = 3.65 days down/year - far less redundancy, nearly double the usable capacity." Which is why the status pages at OpenAI & Anthropic look like a fruit salad of green, yellow and red.
Blue Origin's New Glenn Static Fire Encounters Failure
Blue Origin doing a static fire test of its New Glenn rocket. Something went wrong. video by @spaceflightnow
AI Token Black Market Fuels Chinese Labs’ Data Harvest
Good article on AI tokens black market and how Chinese labs could be getting their training data from US models https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens-in
Claude Team Delivers Live Coding Session at 13:45 PT
Lots of good sessions with Claude team. including a Live coding session with Claude Code and Boris at 13:45 PT
Coinbase Cuts 14% Staff to Become AI‑driven
Coinbase laying off ~14% to rebuild the company "as an intelligence, with humans around the edge aligning it. Non-technical teams are now shipping production code and many of our workflows are being automated."
Anthropic's Edge: AI-Driven Workflow Overhaul Fuels Success
I was pretty sure Anthropic uses Mythos internally, and Boris confirmed that. They also use Opus 4.7 today to write all their code, all their sql... Anthropic's advantage today is that they've reorganized the processes and workflows around AI. If you look...
AI Coding Will Become as Commonplace as Texting.
- Will coding with AI be like a similar skill like using Word or Excel today? Boris Cherny: Yes, it's going to be a skill like sending a text message... Btw, about that printing press take - quite a different path taken...
Claude-Powered AI Agents Run Boris Cherny’s Nonstop Dev Ops
Boris Cherny dev setup: - Claude mobile app is the primary work interface - has 5–10 sessions open at a time, each session can have multiple agents running - runs thousands of agents on deeper tasks overnight - he has dozens of /loops running...
May Sparks Surge of Next‑Gen AI Model Launches
May is going to be a good month for new models. I bet we will see a a few gemini models at Google I/O - Gemini 3.5 [Pro/Flash] Sonnet 4.7 ? gpt-5.6 The hockey stick goes up.
Gemini Reveals Deeper Thinking Traces than Usual
Looks like Gemini showed me a bit more of their thinking traces, than the usual: Tone/Style: Balance empathy with candor, validate emotions (if any - this is mostly factual), ground in fact, gently correct misconceptions. Mirror user's tone (straightforward, slightly informal,...
U
Ukrainian drone still flying after being hit by Russian ground fire. Reminds you of anything?
Jensen Warns AI Scare Tactics Hurt Future Talent
Jensen pushes hard on Dario's narrative about AI destroying jobs. I mean, scaring people, and making AI being perceived as a bad thing - is probably not good for AI business? Jensen: "the things that are said are very counterproductive and,...
Most Media Sites Block Key Bots in 2024
I've been scanning the top 275 media websites [since 2024] for their robots.txt rules and checking which Bots are disallowed. Here is the latest update:
Future AI Will Run Locally, Private, and Fast
I hope we get to SOTA AI models that can be ran locally at useful speeds, currently they choke at around 10tps on my M4 Pro. I believe useful AI models will eventually end up being local, private, yours.
AI Must Prioritize Reliability Over “Move Fast” Mentality
I hope the "move fast and break things" era a lot of AI companies are in right now ends shortly after they IPO and AI becomes critical infrastructure. You would never accept a world where your database sometimes works, sometimes it...