cybercody

A reminder. As with food, we spent most of our history deprived of information and craving it; now we have way too much of it to function and manage its entropy and toxicity.

@far__el @alec_helbling lol, I was going to say is it like dieting or drinking

@alec_helbling More like drunk

Do you think an LLM feels pain when it is quantized?

@ideabrowser lol it’s just different use cases

Markdown is dead?

@AdamKadmon91 @bindureddy Precisely.

@bindureddy If this isn't the one, there will be another. Current models are wasteful and inefficient. And the labs hire engineers not scientists. Someone will crack this, the stuff will all run on your PC, the data centers will be empty, and the comapnies will be bust.

SubQ , a new type of AI model, says they are 50x faster and 20x cheaper than Opus 4.7 and GPT 5.5 In fact, they also say they perform INSANELY WELL on benchmarks and have a 12M context This would be earth shattering, if true - Anthropic/OpenAI's valuation would go to zero 😱

@Fortrancookie @kareem_carr Obvious ≠ stupid.

@kareem_carr "we" You can always spot people who think they're smart but really aren't because they try to speak for "everyone" and not just themselves. Write a compiler and get back to me about how obviously stupid your post is. Even antique spell checkers from the 90s "knew" this.

the revolutionary discovery of ai isn't that computers can be conscious, it's that human languages are a lot more mathematical than we thought

@athenaeumbc Atlantic article link? 🙏

Fewer than half of US adults read a book last year. Even fewer read an actual novel, and the trend is looking worse still for teenagers. Why is nobody talking about this??

@tferriss Excellent newsletter this week! Hoffman’s Faith in the Possible, Dr Tommy Wood’s guest post, and the glymphatic system paper were all very interesting. #5BulletFriday

One of my favorite tools for the past year is Atuin. It's a CLI tool that makes it seamless and easy to find a command you ran in some folder months ago. It's magical. Highly recommend giving it a spin, whether a pro dev or someone new to the shell. atuin.sh

there will this brief era where we can watch our AIs bumble around on the computer clicking things, failing sometimes, taking a ~human amount of time to write code. in the blink of an eye they’ll be manipulating computers far too quickly to monitor

@Timberwolves @rudygobert27 GOOOOOOOD MORNING

GOOOOD MORNING, WOLVES FANS!!! say it back. 🐺

Made a clean and simple Apple Watch app for tracking push ups, Push Up Reps. It was a fun experiment in AI development in a language I do not know. Also learned the ins and outs of Watch development - XCode really doesn't play well with Watch-only apps. So if you ever wondered why all your Watch apps require an iPhone app, it's because Apple makes it tricky to do otherwise. Reply or message me, and I'll give you a promo code. apps.apple.com/us/app/push-up…

@bindureddy Gemini kind of beat me them to the punch? 3B is nice and succinct though!

The big story that everyone missed yesterday - Qwen 3.6 dropped with 3B active params costs nothing to run and delivers 80% of Opus 4.7’s performance 🤯 Open source is making giant leaps

I’ve read both your post and Chase’s, and they aren’t necessarily at odds? Owning and controlling your agent harness and its memory pairs quite well with keeping that harness thin and building a skill flywheel plus resolvers. I’m surprised by your flat “it’s wrong”, seems contrary to your ethos.

@mstockton @j_schottenstein @hwchase17 It’s wrong Thin harness, fat skills: x.com/garrytan/statu…

This post is so so good. We are at this interesting point where we’ve started to really figure out ‘memory’ with these LLM-based systems. And @hwchase17 is totally right - memory is basically just context collected and injected at the right times - it’s probably the most important context, and it *must* be interpretable to you, and portable for you. There is a real danger in adopting systems that won’t allow this for you, or actively prevent you from doing it. I hadn’t put it all together before reading this post, but this is definitely one of the reasons I love building on top of DeepAgents

@alphabatcher It’s so good! It lays out the “why” of current agent architectures very well too.

@cybercody Already quoted this article from Harisson 10/10 to be honest

Great breakdown of how model providers are platformizing their AI/agents. A lot of people will take the convenience of going all in on a provider, but they will be locked in and giving up data control. Literally giving up agency.

@Timberwolves Goooood morrrrning

GOOOOOOOD MORNINGGGGGGG. say it back.

@jmitch @chloepark @xteinknote Anything you like more about the X4?

@chloepark @xteinknote I have both. Prefer weight and size of x3 as well as the button placement

make hardware smol again

Meta dropped TRIBE v2 which maps media to predicted neural activity at a useful resolution. Which is frightening coming from Meta, but also really neat science and includes a neat demo. aidemos.atmeta.com/tribev2