chittem

Building agents @Google. prev - @Oracle, @YourCampusFund. CS @bitspilaniindia

[chittem @AdithyaChittem]

New Paper: mmWave Radar Aware Dual-Conditioned GAN for Speech reconstruction of Signals with Low SNR. 🧵(1/4)

Demo Website with audios and spectrograms: https://t.co/v7X1EobN8y

Paper: https://t.co/DXTi4Wb0CD

[chittem @AdithyaChittem]

@TexasPercy 10/10

[ohshin @ohshinbhat]

i dont understand this codebase anymore man

[chittem @AdithyaChittem]

@sarvagya_kul Was this through backdoor too lmao

[chittem @AdithyaChittem]

@sdianahu Grading reasoning traces is probably one of the hardest things to do considering how much traces vary across model providers and even releases across the same model provider. You'd just have to keep coming up with new measures/evals even if there's a small weight change.

[Xiao Ma @infoxiao]

I just open sourced my “lazy or craftsman?” simple test.

[chittem @AdithyaChittem]

@Ishansharma7390 What the fuck

[sean lee @infinitefun_]

most of what is considered "taste" (read: design) is in the realm of zero sum signaling games

your tasteslop will just be the next AI slop

and then your anti-tasteslop will become the next tasteslop

taste is defined in terms of slop and therefore can never transcend slop.

[chittem @AdithyaChittem]

Is anyone at OpenAI still working on Prism? Was such a great idea I cannot believe they messed it up

Overleaf HAS to go

[maharshi @maharshii]

is it just me or has claude fable 5 gotten noticeably worse?

[David K 🎹 @DavidKPiano]

Too many developers don't understand what "compounding slop" is.

A loop that prompts agents is a great way to automate slop creation. Constrain the state-action space so the loop can't drift, then automate inside it.

Human-in-the-loop = feature, not bottleneck.

[chittem @AdithyaChittem]

AUSTRALIA TIMEEEEEEE LFGGGGGGGGG

[chittem @AdithyaChittem]

New Paper: mmWave Radar Aware Dual-Conditioned GAN for Speech reconstruction of Signals with Low SNR. 🧵(1/4)

Demo Website with audios and spectrograms: https://t.co/v7X1EobN8y

Paper: https://t.co/DXTi4Wb0CD

[K.Bourdieu @infohazard_lol]

Anthropic has seriously shattered my AI psychosis, so now I feel extremely anxious about the sudden lack of perceived hyperproductivity. I don't trust any of my agents, Claude or otherwise, will get a single thing even partially correct. Even for vibe code fun the magic is gone

[chittem @AdithyaChittem]

If a year from now we are still writing 3000 line markdown files we will have completely lost the plot

[Rohan Paul @rohanpaul_ai]

This Meta + Stanford + Illinois survey paper argues that AI agents work better when code becomes their main working layer.

The problem is that an LLM by itself is mostly a text predictor, so long tasks can lose state, hide mistakes, and turn plans into actions in fragile ways.

The real advance is not “AI writes code,” but “AI uses code as the environment it thinks inside.”

The authors call the surrounding system an agent harness, meaning the tools, memory, sandboxes, checks, and feedback loops that turn a model into an agent.

Their core idea is that code should sit at the center of that harness, because code can be run, inspected, checked, saved, edited, and shared.

Tests become sensors.

Repositories become memory.

Logs become history.

Sandboxes become boundaries.

A generated script is no longer merely an answer; it is a handle the system can run, check, revise, share, and roll back.

The main finding is a pattern across many fields: code helps agents reason through executable steps, act through tool calls or control programs, and model environments through tests, traces, logs, repositories, and simulators.

----

Paper Link – arxiv. org/abs/2605.18747

Paper Title: "Code as Agent Harness"

[chittem @AdithyaChittem]

@redrodeo03 @c_engines Congratsss

[chittem @AdithyaChittem]

Sloppification is long to be a large scale problem soon enough and whoever builds smthng that can balance abstraction without missing out on nuance that actually solves issues in production grade code wins no doubt

[chittem @AdithyaChittem]

I do think long term the health of repos everywhere will be near dogshjt. The instant gratification from seeing the sloppy code "work" makes most devs completely ignorant to anything long term. Moreover the increased expectations in terms of output doesn't help this either

[Mario Zechner @badlogicgames]

recommended reading.

https://t.co/2GgZb04PoE

[chittem @AdithyaChittem]

"let's build a review agent" sure that'll report 100 bugs in its report. Will YOU fix it? No.

You feed it back to an agent that does the same thing all over again

[chittem @AdithyaChittem]

@killedbygoogle Never

[Harveen Singh Chadha @HarveenChadha]

I don't know what's more shocking today:

Demis being an angel investor in anthropic
or
Karpathy joining anthropic