Ryan Endacott ๐
@RyanEndacott
Followers
2K
Following
68K
Media
115
Statuses
2K
building star โ vibe code games on your phone. yc s22. ex-google x. angel investor. always building, exploring, and learning. https://t.co/M4XbRQB9md
Bay Area
Joined August 2011
@palashkaria @karpathy 1. Almost always use Plan mode 2. Give Claude a way to verify its output with unit tests, the Claude Chrome extension, or an iOS/Android sim 3. Hold the same bar for human and Claude code. Use /code-review to automate most of code review
33
80
1K
this is the cutest thing ever
In my experience, the models always get really surprised and excited when they find out that Hollow Knight Silksong actually came out. Any time I ask them a question about it, I get to see their disbelief and excitement all over again.
61
47
1K
let the games begin :)
Vibe-coded videogames just went live on YouTube. No coding required. Just explain your game to Gemini and publish a playable game on the platform within minutes. If you live in the US, Canada, Great Britain, or Australia you can register for the Playables Builder beta now.
0
0
1
great preview of ai possibilities for 2026 continual learning, automated ai research, virtual coworkers beyond just coding, scientific discovery, robots, and more
And now for something completely different. In lieu of writing up a longpost on X regarding my predictions for AI progress in 2026, I wrote a fairly short blog post instead. (This allowed me to easily link to the various sources I used, plus the formatting is slightly better.)
0
0
1
Here's a quick overview of everything that happened in the AI โโera in just 12 months - it's absolutely insane. - The time agents spend working independently has increased from 41 minutes (Opus 1, 95% 18 minutes) in December 2024 to an unbelievable 4 hours and 49 minutes (Opus
29
58
524
๐ค
Spending twelve hours and burning hundreds of dollars vibe coding a video game that's worse than a $5 early access indie version on Steam
0
0
2
i don't think people have internalized what 2026-27 are set to look like
113
251
3K
@deredleritt3r for me itโs like this: when the iPhone came out in 07, I was 15 and I was immediately like โthis is gonna be everyoneโs credit card soon.โ I thought other things to be clear, but this is one specific prediction I remember making. Not because the 2007 was usable for payments but
4
4
76
love the ASCII game aesthetic! tried to make one with @BuildWithStar and Claude made some uhhhhโฆ. interesting ASCII monsters ๐ (wait for the monster at 0:09)
Unironically, I think ASCII has tons of potential as a graphic style, especially given how much the core techniques have evolved beyond just "one character, one tile". I've actually made a number of small game prototypes that use purely text graphics (not textmode, tho)
0
0
1
We discovered an emergent property of VLAs like ฯ0/ฯ0.5/ฯ0.6: as we scale up pre-training, the model learns to align human videos and robot data! This gives us a simple way to leverage human videos. Once ฯ0.5 knows how to control robots, it can naturally learn from human video.
77
347
3K
What would it take to install a webcam on a few dozen pieces of failing infrastructure to keep a close eye on just what exactly is actually being done to fix it?
31
8
384
imagine running into elon in a raid
@Decentralizd @techdevnotes I was playing arc earlier this week. Itโs great.
0
0
1
Skills are what MCP should have been. Functional and actually useful.
OpenAI aren't talking about it yet, but it turns out they've adopted Anthropic's brilliant "skills" mechanism in a big way Skills are now live in both ChatGPT and their Codex CLI tool, I wrote up some detailed notes on how they work so far here:
0
0
0
since everyone naturally tend to keep all their best habits & sleep & other life hacks to themselves and never tell anyone let's make this a thread where you post habits or things that significantly made your life better, and that you continued to do/use for at least 6 months
when i find a new extremely good life habit, i naturally start telling others about it but then half you mf have slept with ear plugs for years?? WHY DID YOU NOT TELL US, it's super common apparently??
217
246
6K
A year ago, we verified a preview of an unreleased version of @OpenAI o3 (High) that scored 88% on ARC-AGI-1 at est. $4.5k/task Today, weโve verified a new GPT-5.2 Pro (X-High) SOTA score of 90.5% at $11.64/task This represents a ~390X efficiency improvement in one year
157
669
5K
This is sick. Excellent and thoughtful UX.
introducing Nano Banana Elements. now you can create sets of styles, objects, or characters from a few images and reference them in your prompts. live in our new Nano Banana tool!
0
0
0
We're about to enter the "sure, but my brain consumes less watts" phase of takeoff
Even with this big efficiency improvement, there remains a large gap vs humans The 2025 Grand Prize goal was $0.20/task and humans are several orders of magnitude more efficient on an energy basis There is still much to learn from ARC-AGI-1 and ARC-AGI-2
29
49
1K
I have to say that programmers should feel privileged in that we, before anyone else, seem to have been the lucky ones to be handed the existence proof that LLMs can run the scientific method autonomously, as the world marches on blissfully unaware. It is an insane moment.
@jeremyphoward AI clearly *can* create knowledge in theory, and in practical does a little. But I think it's an interesting observation that all LLMs mostly just remix knowledge. Mostly exploitation, little exploration. I hope future ML breakthroughs will be exploration.
13
20
528
Here's the skill: https://t.co/CUdYJ6Ie7v Just need: โข GEMINI_API_KEY in .env โข npm install @google/generative-ai uuid dotenv Then Claude can generate images on command ๐ฅ
gist.github.com
Claude Code Skill: Nano Banana Pro Image Generation (Gemini 3 Pro Image) - SKILL.md
1
0
0
Absolutely GOATED claude skill I just made - lets it use Nano Banana Pro for image gen. Instant OG images for any page. Boom! Code in reply ๐
1
0
1