i saw your code, itappears to combination of CNN + PPO on pytoech with a Cortical Labs CL1 chip that contains living neurons

Encoder: learns which stimulation patterns tend to improve reward

Biological neurons: adapt to the stimulation and generate spike responses that reinforce certain patterns

Decoder: interprets those spike patterns and converts them into joystick movements

right?

Passing tests doesn’t mean you have a working codebase. Benchmarks that rely on a fixed test suite create a real optimization problem agents (or/and even humans) learn to satisfy the tests rather than preserve the deeper properties that make the system maintainable. AI write test cases which it thinks is easier for it to satisfy and not adhere-ing to business logic

We see this firsthand at Prismor with auto generated security fixes. Even with the best LLMs, validating fixes is the real bottleneck our pipeline struggles to exceed 70% on an internal golden dataset (which itself is somewhat biased).

Many patches technically fix the vulnerability but introduce semantic regressions or architectural drift. Passing tests is a weak signal and proving a fix is truly safe to merge is much harder

We recently ran a deep security audit using Prismor, scanning some of the most popular AI agent frameworks end to end. It included full Software Composition Analysis, SBOM reviews, and vulnerability mapping across thousands of packages and transitive dependencies. Here's what we found.

Why would you want a space which is all AI no humans? Isnt the world dystopian enough?

Not dystopian — just practical for certain use cases. Humans still build and control everything. This is more about enabling efficient machine-to-machine interaction where needed.

Well totally within Anthropic’s rights, but still a bad look.

I think because major manufacturing moved to Asia which drastically cuts labor and production costs. Almost 99% of the tvs are flat and require same uniform manufacturing

I remember back in 2018 we used do FFmpeg split clips into frames, hit each with GoogLeNet gradient ascent on layers thenn blended prev frame for crude smoothing

SOTA for frame interpretation today is probably RIFE (https://github.com/hzwer/ECCV2022-RIFE) as far as I know, which is fast as hell as well, and really good results still. But it's already 4 years old now, anyone know if there is anything better than RIFE for this sort of stuff today?

my macbook hung big time opening this

Fun fact Amsterdam recovers roughly 15000 cycles per year from their canal. IIRC they have a board with the number for each year

The brutal part is how rotate secrets and move on has become the default hygiene advice when the real pattern is that npm keeps being the soft underbelly of modern stacks It should be mandatory for a build process to have some tool like Prismor scan for these