Arno Khachatourian

I made a searchable directory for @karpathy's YouTube tutorials, with quizzes for each video. Links below:

@Teknium They can always add more query params later

Humans& missed the opportunity to named themselves Her&

@steipete The bot formerly known as molt

Kudos to Wired for a great article! I will never get rid of the 2-day molt part of the renaming saga will I

@vikhyatk @_mattneary Much too Byzantine

@_mattneary TLA+ finally going to get popular??!

you don't have to write code. you don't even need to review every single line. but you have to understand every critical part but how do you develop judgement of what matters, without getting burned by mistakes you made, without ripping your hair out debugging impossible bugs?

@littmath ah, yes, my favorite type of wrong. "being too early" gets an honorable mention.

learning a very useful euphemism for “wrong,” namely: “directionally correct”

@ryanlpeterman Always look forward to your videos! You may want to set up the "Notify me" thing on YT. Not quite sure how that feature works, though. Complete aside, but I just recently set up an AI summary page for your interviews if you want to check it out: learngood.com/#/youtube-seri…

New grad to Principal Eng (IC8) at Meta career story launching Monday. Some interesting bits from the conversation: 1. Concrete example of not doing the team's work in favor of a calculated risk to start something new which paid off (and he would've done if it didn't pay off) 2. How to maximize your luck when it comes to networking / people opportunities 3. How Principal eng (IC8) promos differ from lower level promotions and how to work towards the promo in advance of the promo committee We went over each of his promotions, the stories behind them, and also a bunch of other interesting stuff because of the domain he was in If you're interested you can follow along either on YouTube, Spotify, or Apple Podcasts to get it when it drops! YouTube: youtube.com/ryanlpeterman Spotify: open.spotify.com/show/0MX9PyeCz… Apple Podcasts: podcasts.apple.com/au/podcast/the…

@henrytdowling @swyx And you get some bonus takeaways as well. Overall, there’s prob a lot of variance on how useful these things are.

@henrytdowling @swyx It’s qualitatively different. I consider it an alternative to clicking on the video and reading the YouTube comments to see what it’s about (rather than considering it an alternative for watching the video). It helps me decide whether to watch or not, which is rare for me

I generated some summaries of the recent AI Engineer videos (learngood.com/#/youtube-seri…). Far from perfect, but hopefully useful. You can download anything you’re interested in as markdown and use it in context with your fav LLM. Cheers @swyx! The slop is served.

@henrytdowling @swyx Hard to say. If I have the video playing in a separate tab I barely absorb anything, and reading a summary at least gives me something, but if it’s a talk I’m really interested in and watch undistracted, it provides way more value than reading a summary, maybe 3-10x

@arnokha @swyx Genuinely curious- what % of the content do you think you're getting from these summaries vs watching the vids? Have been playing away with bulk summarizing youtube video essays as well

people are telling me that my shaman (opencode) was the wrong one. Apparently the shaman down the street (Claude code) had the real incantations Will try again, but with CC

@bcherny @Bhoumik03 I think that there can be a quicker win by specifying something should survive compaction. Maybe like `/remember $` Where that specific message survives and is fed into the prompt at compaction time. Should probably also be displayed to the user in case they themselves forgot

@bcherny @Bhoumik03 Maybe outside the scope of just /compact, but maintaining a living map of filename -> (description, hash) and feature -> (description, relevant_files) across CC sessions would be awesome

👋 Hi I'm Boris and I work on Claude Code. I am going to start being more active here on X, since there are a lot of AI and coding related convos happening here. Feel free to tag me with Claude Code feedback or bug reports. Love to hear how y'all are using Claude Code, and what we can do to make it even better.

@bcherny @Bhoumik03 I've since tried to hack a solution to this for my own test harness, and now think it's probably too much complexity for what it's worth.

@__paleologo Bobby Eggers doesn't miss. His movies are so dreamlike, captivating, and heavy. It feels like I’m watching movies that were made by Jung, featuring Willem Dafoe, and I love it.

PSA: "The Northman" is streaming on Netflix. Every movie by Eggers is a masterpiece of sorts and a retelling of some myth. Maybe "The Iliad" would be more suited to be directed by him than Nolan. But ok.

@thdxr Have you found a good way to eval/optimize this kind of thing, or is it just you and the team noticing and intuiting prompt/workflow improvements for now?

in opencode 1.0.203 we shipped some prompt tweaks for the bash tool it was using bash way too often to search instead of grep/glob tool and when it did it would often do overly restrictive searches and miss files it should be paying attention to simple fix, big improvement

@andrew_n_carr Claude can’t even believe the current version of Claude exists

claude cannot believe there is a sam 3. it wants to replace everything with sam2-hiera-large

@NitinSub do you use something like this, also? alias l="ls -tho"

there are few function in my ~/.zshrc that have saved years of my life this is one of them

@NitinSub lol no way i use this too qp() { COMMIT_MESSAGE=${1:-"quickpush"} git add . git commit -m "$COMMIT_MESSAGE" git push }

@Teknium nano banana vs openAI on same prompt and base image. Idk which is better, but john martin remains goated.

Tried chatgpts new imagen.

@Teknium @OfficialLoganK @TheCodeOfJoel would you consider releasing it at some point?

@arnokha @OfficialLoganK why is everyone holding back their pokemon harnesses

Gemini 3 Pro playing Pokémon vs 2.5 Pro (we used to all be impressed by 2.5 Pro)