Isn’t this non self in Buddhism ?

This might be dumb but I don’t know what it would mean to exist in the sense you’re describing.

The context is a state

I feel like a transformer with tools that responds to external outputs is no longer just a big matrix multiplication. If it’s high dimensional enough.

Doesn’t this break down if the model is being trained

GitHub - alibaba/OpenSandbox: OpenSandbox is a general-purpose sandbox platform for AI applications, offering multi-language SDKs, unified sandbox APIs, and Docker/Kubernetes runtimes for scenarios li... OpenSandbox is a general-purpose sandbox platform for AI applications, offering multi-language SDKs, unified sandbox APIs, and Docker/Kubernetes runtimes for scenarios like Coding Agents, GUI Agent...

Alibaba's OpenSandbox

It is a general-purpose sandbox platform for AI applications, offering multi-language SDKs, unified sandbox APIs, and Docker/Kubernetes runtimes for scenarios like Coding Agents, GUI Agents, Agent Evaluation, AI Code Execution, and RL Training.

github.com/alibaba/Open...

My manager tells me the same thing

If you work in tech, and are ever approached by any crypto sh*tcoin even for a harmless ask ("here's some money for you/your project! Take it! No strings attached, promise!"), a warning:

Talked with someone who took it.

Seemed too good to be true.

Death threats to them and family followed.

It’s not perfect ofc. But the amount of code I write by hand is decreasing rapidly. I think it will hit less than 10 % this year.

I have literally wanted this feature for years and @mariaa.bsky.social threw it together!

Beaker Next - Dev Diary #6 - Building a social feed application YouTube video by Paul Frazee

If you enjoyed @danabra.mov's social filesystem blogpost, you might get a kick out of a literal exploration of the premise I tried using the dat p2p filesystem 6 years ago www.youtube.com/watch?v=k0JV...

My bet is that AI policies will be soon avoided by most companies / projects in favor of quality centric policies.

I think AI allows us to have some new definitions of quality. Token consumption, error rates from something non human.

Who will verify the verifiers

Only the next quarters stock price matters :)

There will still be devs. What’s not clear is if we need fewer of them. Current trends indicate much weaker junior and new grad hiring.

Nice idea

Dump lots of stuff into context using voice

Some tricks -
Use the names of famous authors you like as prompts.
Use multiple models - Gemini , Opus , Kimi
Stick to word limits
Have prompts that compile and flag common slop patterns
Simulate a conversation on the topic between different characters

Il trucco DECISIVO!!111 per lavorare coi coding agent YouTube video by Salvatore Sanfilippo

My latest video: the only SECRET!!111 you need to know to work well with coding agents.

www.youtube.com/watch?v=TJ6r...

Make sure to check lobste.rs comments on my blog post, if you are interested in the topic of AI coding. They are better than HN comments about the same post.

New blog post: Don't fall into the anti-AI hype.

antirez.com/news/158

ok i've tried a few of these, and @bobbby.online's Deciduous (notactuallytreyanastasio.github.io/deciduous/tu...) is most intriguing so far

it roughly matches what i want out of a tool like this

not perfect but i recommend playing with it

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning…

I updated my list of reasoning models with solid tech reports for 2025!
2025-01-22 - DeepSeek R1 - arxiv.org/abs/2501.12948
2025-01-22 - Kimi 1.5 - arxiv.org/abs/2501.12599
2025-03-31 - Open-Reasoner-Zero - arxiv.org/abs/2503.24290
2025-04-10 - Seed-Thinking 1.5 - arxiv.org/abs/2504.13914

Good to see AI discussions picking up here.

I’m glad we can discuss AI on here now without getting cancelled

Use hooks as life cycle events to inject reminders / context.
Orchestrator spawns jobs and runs additional checks.
Skills work the same.
Subagents work the same.

Look into Claude SDK and write your own

might be time