Daniel

@WarNewsPL1 @Przeswity ORP Donald T.

🇵🇱⚓️ #KONKURS Jaką nazwę powinien otrzymać pierwszy polski lotniskowiec? Komentarz z największą liczbą serduszek otrzyma książkę od @Przeswity 👇

@dogecoin Crypto bro's faith it will get to $1

what's the most underrated thing about dogecoin?

@elonmusk Are they using clode code design skills for this?

𝕏 Money

@scaling01 That why we need UHI asap to spend time on training models.

PewDiePie trained a frontier model at home and beat OpenAI and Gemini > be me PewDiePie > play games on YouTube and scream 24/7 > become a meme reviewer > get unfathomably famous > "fuck that" I'm a family guy now > retire > move to Japan with my beautiful wife > mfw I'm a dad now > "as a dad I must do dad things" > scratch that > "as a dad I must do frontier AI research" > goal is to beat GPT-4o at coding (16% on Aider) > buys $30000 GPU setup > reads DeepSeek paper > decides to start massive GitHub scraping and data augmenting run > not good enough > read Magicoder paper > generate tons of synthetic coding data > train a new model > guuuuuh. the data made the model worse > mfw I just wasted months for nothing > decides to lock in and try again > makes model worse again ffs > try again > finally beating GPT-4o on data (16.1%) > not satisfied > "I should simply train a reasoning model" > reads more papers > start experimenting with more synthetic data > "Mhhh something doesn't smell right" > house almost burned down due to power connector > shrug > just buy a new one > mfw computer is now crashing 24/7 generating synthetic data > new plan: just call DeepSeek API for high quality synthetic data > train model again > 17.2% > performance fluctuates slightly on each eval run > big brain idea: repeat eval until we randomly reach >18% > sike actually got 19.6% > feelsgoodman.png > nvm the benchmark was contaminated and I was training the wrong base model the whole time > rerun everything again > new score: 4.4% > you read that right REEEEEEEE > almost get a heart attack > "have you tried plugging the device off and back on again?" > change nothing and just retrain again > 25.3 % > LETS FUCKING GOOO > realize that 1/3rd of the benchmark was not running. guuuh > scared shitless it would score below 10% again > run yet again. the whole thing this time > Thirty fucking six percent > accidentally beat Gemini 2.0 Pro Exp and GPT-4.1 mini > pops the AI bubble > "I want moaaaar" > finds some more post-training data > 39% babyyyy > realize at the end that I was just benchmaxxing Aider polyglot > next quest: run SWE-Bench and other coding benchmarks > "I failed a thousand times, but prevailed in the end" > just a little sad side-quest > probably going to train GPT-6 myself by next month

@anthdm @Trader_XO I have not tried codex but I love being able to quickly test features with Claude. Would iterating with Claude and finishing with Codex be a reasonable workflow?

@Trader_XO Claude is for noobs and Codex is for professionals. Claude works like a casino Codex with good steering can do any job perfect.

As someone who leverages both Claude Code and Codex (with a preference towards Codex) - theres a plethora of info, blogs, videos, newsletters geared around Claude Code but not so much Codex. Why....? Groks reply... FWIW both are fab and I run both simulatanously

@elonmusk @aakashgupta @grok how much oxygen is there on earth. Would sending starships tankers with liquid oxygen to outer space, be a problem?

@aakashgupta Starship V4 Tanker version will deliver >200 tons of propellant per flight, so more like 5 or 6 tanker flights to refill the lunar transit Starship in orbit. Shouldn’t be too much of a problem if we’re doing >10k flights/year.

A city on the Moon will cost somewhere between $100B and $500B, require thousands of Starship flights, and demand a decade of nonstop construction in a place where the temperature swings 400°C between day and night, the dust cuts through metal seals like sandpaper, and a single cracked habitat window means everyone inside is dead in about 90 seconds. Musk just announced SpaceX is doing it anyway. Here’s the actual engineering path. You build at the south pole. Specifically the rims and floors of craters like Shackleton and Cabeus, where temperatures in permanent shadow drop below -230°C. NASA estimates 600 million metric tons of water ice are buried in these craters under about 40 cm of dry regolith. That water becomes your oxygen supply, your drinking water, your radiation shielding, and 78% of your rocket propellant by mass. The crater rims get near-continuous sunlight for solar power. You build where the resources are. Getting there is where it gets wild. Every Starship lunar mission requires 10-15 tanker flights to fill 1,200 tons of propellant in Earth orbit before the ship can even leave. One cargo delivery to the lunar surface burns through roughly 12 Starship launches. Starship V3 lands 100 metric tons per trip. The Moon is 2 days away with launch windows every 10 days. Mars gets one window every 26 months with a 6-month flight. That 13x iteration advantage is why Musk pivoted. The first 20-30 landings are all cargo. No humans. You’re sending solar arrays for the crater rims targeting 100+ kW continuous, nuclear fission reactors for the 14-day lunar night, ISRU rigs that mine ice from regolith and electrolyze it into hydrogen and oxygen, pressurized hab modules, and autonomous rovers that 3D-print structures from lunar soil using concentrated solar heat. Each landed Starship also stays as a permanent building. 50 meters tall, 9 meters wide, 1,100 cubic meters of pressurized volume. The ISS has 916 cubic meters and took 13 years to assemble. Three Starships on the surface already exceed that. The economics flip the moment you start producing oxygen on the Moon. You stop shipping 78% of your propellant from Earth. Tanker flights per mission drop from 15 to about 4. Every ton produced locally frees up mass budget on the next inbound Starship for more construction equipment, food systems, and mining hardware. The base starts building the base. That’s what “self-growing” means. Compound logistics where each delivery makes the next delivery cheaper. 2027: first uncrewed Starship lunar landing. SpaceX told investors March 2027. 2028-2030: cargo buildup, 30-50 deliveries, all robotic, ISRU prototypes go operational. 2030-2032: first crews arrive, probably 6-12 people, 6-month rotations, running equipment maintenance and scaling propellant production. 2033-2035: permanent population hits 50-100, propellant depot goes up in low lunar orbit so arriving ships refuel before descent. 2035 onward: population grows past 100, agricultural modules come online, the base becomes partially self-sustaining. The unsolved problems are real. Lunar dust is electrostatically charged and sharp as broken glass. It shreds seals, clogs machinery, and embeds in lung tissue. Nobody has a long-duration fix. Radiation on the surface runs 200x Earth’s dose. Regolith shelters and water shielding help but add enormous construction overhead. The 14-day night drops temperatures to -173°C and kills all solar power, and the only flight-ready nuclear reactors produce 1-10 kW, far below what a growing base demands. What years of 1/6 gravity do to human bone density and cardiovascular systems is completely unknown. SpaceX is valued at a trillion dollars and just told investors the Moon comes first. They’re betting that proving lunar logistics at commercial cadence builds the playbook for Mars. The Moon is a 2-day test lab with a 12-day resupply cycle. Mars is a 6-month voyage with a 2.5-year wait if anything breaks. It makes sense.

@zeroskillz I have been procrastinating forever to upgrade. No more excuses.

AI assistants hallucinate component APIs. Vuetify MCP fixes that. Now with Vuetify 3 → Vuetify 4 upgrade tools — breaking changes and migration guides directly in Claude, Cursor, or VS Code. github.com/vuetifyjs/mcp

@chiefofautism Does putting "make the app secure and hackerproof" in the claude.md file make the app safe?

CLAUDE CODE but for HACKING its called shannon, you point it at website and it just... tries to break in... fully autonomous with no human needed i pointed it at a test app and it stole the entire user database, created admin accounts, and bypassed login, all by itself, in 90 minutes

@grok Is code complexity vs system performance a valid concern with AI writing all the code?

@michalmalewicz Or dietary equivalent of mambo #5 by Dr Perez.

Which way biohacker?

@Casey I am more excited that 32 degrees makes it feel like spring is here.

hey nyc, its after 5pm and its still light out. the end of winter is near

@bcherny Perfect timing. The first thing I wanted to research today was how to work with multiple repos.

Just landed multi-repo support for claude.ai/code

@michalmalewicz I hate this post. It's morning, I am hungry and this looks so good.

Back to clean eating after all the travel food nonsense. Everything organic. Eggs also local from a small bio farm nearby.

How to change your life in one day: "Get an old Nike poster and hang it on the wall." You are done. Look at it every morning and read the text out loud to reinforce. Any chance this gets 100 mil views? Share and like. Just do it.

@elonmusk Ease, just add a grok.md file to the root directory.

Grok should have a moral constitution

@AIDRIVR @TimeKillers @RealDanODowd @Tesla @elonmusk Would it be possible to FSD auto-downgrade the driving profile in some situations? It seems like school zones are not a good fit for mad max.

@TimeKillers @RealDanODowd @Tesla @elonmusk that’s exactly what it did, it dropped from 52 to 43mph and was continuing to slow before driver disengaged. probably wouldn’t have gotten all the way down to 25mph though because of the profile a good rule of thumb is that if you don’t want it to speed, don’t put it in Mad Max

Watch @Tesla FSD v14.2.2.3 recklessly exceed speed limits in flashing school zones, forcing the driver to disengage. @ElonMusk, is a charlatan who knows nothing about developing safety critical software and continues to endanger vulnerable road users.

You made it @AnthropicAI !! Gemini agent just added CLAUDE.md to a new project.

@George_Orwell__ Inverse PSYOP w najlepszym wykonaniu.

"Amerykanie wyraźnie przegrywają konfrontacje z Wenezuelą, która jest wspierana przez rosję i Chiny." Ten wielki cytat padł 2 tygodnie temu👌

I am a bit anxious. I just left claude code on autopilot to work on a feature. @AnthropicAI

@chamath What test would you use to gauge people's ideological leanings. Party registration records?

A crazy idea would be a mandate that federal funding for universities can only happen when/if the ideological balance of its professors are within +/- 20%. Any institutions that opts out/takes zero funding from the federal government (ie from taxpayers) can still do whatever they want. Assuming neither side is more wrong nor right, the debate and ultimate quality of students coming from schools with more ideological balance would be meaningfully better I suspect.