crorella

Should be both the entity providing the platform where the inference is running + the entity giving the prompts/instructions.

It looks like a lot like how data bases query optimizers work, with the exception that in the paper there is also a learning/memory component that conditions the evaluation of the answer provided by the first model.

What are good options to run local models nowadays? Something good for coding and personal assistant kind of things

This sound very interesting, do you have any resource I could look at? Me and my son did a very rudimentary (compared to yours) setup to play Paranoia, but this is at another level.

it has been for ~2 years already, it is like voice mode and you can talk for up to 30 mins with it IIRC

Fun/interesting to see how opensource models surpassed Anthropic's

I had the same visceral reaction lol, so ugly.

Tokenmaxxing makes no sense, it is akin to write extremely inefficient SQL / Spark Jobs, full of cartesian joins, ultra skewed datasets, etc, just for the sake of using as much compute / memory / IO as possible.

This always happens when the metric becomes the goal, companies should nurture and foster an environment where AI is used in the most efficient way possible, first asking "do we really need an agent for this" and if so, what kind of agent is needed, what model, reasoning level, etc.

They should also promote projects that aim at saving tokens, increasing cache hits, codifying the information in ways such they use as less context as possible (graphs of knowledge are pretty good for this!)

what a weird surface to put LLMs

Same! At first I was wary of using it because the UI looked less polished, but from the start the stability has been vastly superior and now the UI is much better too.

as a bonus, I have a old version Emby Theather (the windows form based one) that plays 4K with no issues on my computer unlike browsers that fail at that.

wouldn't move to local models in the future remove part of that risk for companies?

underrated comment, this is going to be the main differentiator going forward, the more powerful and versatile harness the more the models will be able to achieve and better/more advanced products will come out of it.

* Parallel agents: Run multiple coding tasks at once so you can move faster across projects

* Isolated workspaces: Each task runs in its own worktree so changes stay organized and conflict-free

* Review changes: Inspect diffs and proposed edits before applying them to your codebase

* Stay in flow: Switch between tasks without losing context or progress

* Automate workflows: Reuse repeatable tasks with configurable skills

* Works with your tools: Open changes in your editor and integrate with your existing setup

* Track progress: Follow what each agent is doing in real time

* Windows sandboxing: Run agent mode inside an OS-level sandbox, with bounded permissions and explicit approval for escalation

* Ship faster: Move from idea to working code with less overhead

I am sad to know about this, Dan Simmons had a mind blowing amount of imagination and the ability to turn that into interesting and imaginative books that expanded my imagination when I read them.

I loved Hyperion cantos, Illium and then non sci-fi books like A Winter Haunting and Summer of night (which I read in the wrong order lol).

I am also happy to read that he was a great person overall and a great teacher. May he rest in peace.

Welcome :D

The preprint: https://arxiv.org/abs/2602.12176

The variety of tasks they can do and will be asked to do is too wide and dissimilar, it will be very hard to have a transversal measurement, at most we will have area specific consensus that model X or Y is better, it is like saying one person is the best coder at everything, that does not exist.

The thrill of competition

Same here! I think it would be good if this could be made by default by the tooling. I've seen others using SQL for the same and even the proposal for a succinct way of representing this handoff data in the most compact way.

what I had in mind when I added that comment was for coding, with the use of .md files. For the web version of chats I agree there is little control on how to tailor the way you want the agent to behave, unless you give a initial "setup" prompt.

It’s like having 3 coins and users preferring one or the other when tossing it because one coin gives consistently more heads (or tails) than the other coin.

What is better is to build a good set of rules and stick to one and then refine those rules over time as you get more experience using the tool or if the tool evolves and digress from the results you expect.

At this rate, in a few months we will have probably some high quality shorts entirely generated by this.

Personal experience here in a FAANG, there has been a considerable increase in: 1. Teams exploring how to leverage LLMs for coding. 2. Teams/orgs that already standardized some of the processes to work with LLMs (MCP servers, standardized the creation of the agents.md files, etc) 3. Teams actively using it for coding new features, documenting code, increasing test coverage, using it for code reviews etc.

Again, personal, experience, but in my team ~40-50% of the PRs are generated by Codex.

same, I had a great idea (and a decently detailed plan) to improve an open source project, but never had the time and willpower to dive into the code, with codex it was one night to set it up and then slowing implementing every step of what I had originally planned.

omg, this is something I've had in mind for quite some time, I even bought some i2s devices to test it out. Do you have some pointers on how to do it?

I saw an "Exalted engineer" once, not kidding.

Insurance and health insurance companies must be super interested in this research and its applications.

a recruiter told me it is false and that they will release some comms soon.

this is false :(

We use this! With BirdNET-Pi gifted some to a couple of friends too!

We configured it so it send me real time alerts when uncommon birds are detected.