Idan Beck

Building autonomous AI systems - @zerg_ai

[Idan Beck @idanbeck]

also... I really want this lego set

[Idan Beck @idanbeck]

@EthanEvansVP I think everyone hates that book, but it's also a great book that everyone should read

I once gave it at a company white-elephant, the recipient was very disappointed

[Idan Beck @idanbeck]

@tenderizzation just check out this bad boy https://t.co/8KodJEFcpV

[Idan Beck @idanbeck]

@endel something similar finally got me to take the red-pill https://t.co/2f5uNqHJtZ

lol was worth not seeing my kids for 3 days straight

[Idan Beck @idanbeck]

@tiago_aoalv I've been using AI to code-gen for FPGA - just not the most sharable thing...

[Idan Beck @idanbeck]

You could solve public transit in CA for under $40K

[Idan Beck @idanbeck]

I have lived among the LLMs, I slept and ate like them, living each moment one prompt at a time

[Edward Lando @edwardlando]

Like many VCs, I really do believe Anthropic and OpenAI are the last startups.

There is no point to building anything else.

Everything has been solved or will be shortly.

[Idan Beck @idanbeck]

@p1nosaur hey but why

[Idan Beck @idanbeck]

Have been using Zerg/AI to revive the Dream engine I wrote in mid 2010s

[Idan Beck @idanbeck]

damnit codex I said it was a nothing burger ONE time - it's all you talk about now

[Idan Beck @idanbeck]

@thsottiaux We would use that a lot for @zerg_ai

[Idan Beck @idanbeck]

@HotForMoot This is the way

[Idan Beck @idanbeck]

At @zerg_ai we are constantly thinking about how software/technology will change with the different speed limits that govern current capabilities, for example:

- token speed
- context size

A short essay on the subject of the latter -

[Zerg AI @zerg_ai]

1/ 🧠 A 1M token context window changed what an agent could do in one call.

When a 1B token window arrives at scale, it will change what an agent can hold in its head.

[Idan Beck @idanbeck]

new PB fam

[Idan Beck @idanbeck]

It is painful to see how little attention this is getting - this is really awesome - have yet to see "world model" models actually show user/environment interactions like this - looks like you're utilizing an underlying deterministic engine with a top level video-gen model? Have always thought this was the golden path since GameGAN / GANCraft - equivalent of neuro-symbolic program synthesis for world modeling.

Very cool work

[Idan Beck @idanbeck]

@Saint_n0mad got me with this thirst trap

[Idan Beck @idanbeck]

it's interesting how quickly a 1M context doesn't feel all that big when it was not that long ago that 32K felt huge

[Idan Beck @idanbeck]

Fun stuff I find around my parents house on the weekends

[Zerg AI @zerg_ai]

Most AI coding agents write code by retrying — same broken approach, slightly rephrased.

A new paper from CMU, U Washington, and Arm asked the obvious question: What if the agent actually *remembered* what went wrong?

🧵