Sabitlenmiş Tweet
Kyle Noble
132 posts
Kyle Noble
@KyleNoble
pirate shipping at @roborobots
Pasadena Katılım Eylül 2010
423 Takip Edilen229 Takipçiler
time to invent offline languages
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
The last 20 minutes of this is pure magic. Fly on the wall while peak talent creates
youtu.be/YgtfHLUIggU?t=…
YouTube
English
@youwillmakemaps Impossible not to do good work in a place like this
English
@realmcore_ Scuffed copy pasting but putting it in a Claude skill now that it seems to be replicating to more tasks.
English
@KyleNoble Isn't it incredible? how did you set yours up? Claude + delegate?
English
@pzakin Need a markdown only server / browser. Skip HTML token excess all together.
English
I'm always intrigued by products that reduce the size of an atomic unit.
Given ^^, I do find myself wondering if -- in the age of vibe coded software -- we see rising interest in single-page HTML docs.
AI-generated HTML docs can take you extremely far.
English
Token efficiency is going to be a very popular sales pitch in 2026
kepano@kepano
It's not because your agent can do something with 12 tool calls and 20k tokens that it should. For example, Obsdian CLI has a command called "orphans" that finds files with no incoming links. Now your agent can get a deterministic answer in 1 millisecond instead of 4 minutes.
English
Co-sign. I’d add, though, that the best version of this isn’t built for humans. It should just be an MCP plus good cli tooling.
Joe Harris@_joe_harris_
2026 is a terrible time to build a robotics company. Imagine building a software company without AWS, Stripe, or GitHub. Everything from scratch. That's robotics today. Most robotics teams waste their first 6–12 months rebuilding the same data infrastructure, telemetry, internal tools, pipelines. None of it is core IP. I started @AlloyRobotics to SPEEDRUN all of it. A modern stack so robotics teams can iterate 10x faster than their competition. If you're building in this space, lets talk.
English
due to a shortage of beds, my catholic mother is being forced to allow my girlfriend and i to share a bed this christmas.
there is, however, a catch
English
Too many people talking about being GPU poor. Not enough people talking about being SPS poor
English
@realmcore_ Ah yah. The ones with the highest number of stars on that list are all fairly complex. Could be a good test set
English
@KyleNoble Oh yeah I understood that, was more so wondering what your definition of a moderate complexity commercial app would be, but having a list right there helps too!
English
Genuinely though guys, please help me come up with tasks.
I'm super curious, what are the dev tasks that IF you could get an agent to do, you would be convinced agi is here
akira@realmcore_
Currently building out our eval stack Need some cool, complex projects like this one to test on:
English
@realmcore_ oh, I actually meant the reverse, taking a react native app and splitting it out into a swift/swiftui app and a kotlin/android app. Could take any of the more complex ones from this list: github.com/ReactNativeNew…
English
@KyleNoble What is an example app that if built in react native would qualify for you?
English
@pratyushbuddiga Darkness at Noon - Arthur Koestler
Dance for Two - Alan Lightman
Norwegian Wood - Haruki Murakami
English
Need book recommendations - feel like I’ve fallen out of love with reading last few months and need something great to get the motion back.
Here’s the embarrassingly short list of what I’ve read so far this year so far - open to stuff similar or totally different.
English
@realmcore_ No problem and totally agree. Auth is a massive issue for agents (and humans too)
English
@KyleNoble Sick thanks! I think there is so much in this space that is completely and utterly unexplored and will need to be for any form of deeper adoption to happen otherwise the friction will be so high
English
@bgurley @brian_armstrong I would expect reasonable disclosure guidelines from regulators at some point as well as the token markets beginning to incorporate data transparency into pricing/valuation.
English
@brian_armstrong The only counter I would make is many private companies prefer to keep their own data private (non transparent). Two way street.
English
Every entrepreneur I know thinks the fundraising process is broken.
Why capital formation should be onchain:
- It’s more efficient
- The process is fairer for both sides
- Much more transparency
English
@willllliam @gpudad can probably answer most of your questions. He’s pretty deep in the market
English
one thing I believe will be very clear in retrospect as browsers evolve:
we will have a consumer PERSONAL browser (optimized with our personal memory, our credit card, our social graph, buying history, and the many agents and apps of daily life)
and we will have a WORK browser (optimized for teamwork via graph of who we work with and permissioning, working across apps, tapping enterprise memory, agents as colleagues etc).
and “browser” will probably become an antiquated term as this interface becomes the OS itself (or at least far more relevant!)
English