Anoni Mouse

Aging corporate engineer dreaming of building something I believe in one day.
Linux🐧 NixOS❄️

[Anoni Mouse @__AnoniMouse__]

https://t.co/QQ4GfCgWo7

[Anoni Mouse @__AnoniMouse__]

@francip If anyone starts using generative AI to run weapon systems I'll be very fucking scared. Even as an ally - the friendly fire we'll take 😅☠️

[Anoni Mouse @__AnoniMouse__]

@ImLunaHey Both are great. They're not mutually exclusive imo.

Sometimes I want to have complete control and to craft everything, other times I want to move quickly.

[Anoni Mouse @__AnoniMouse__]

@jun_song Absolutely. It's about efficiency - not just chucking more and more compute at the problem.

OpenAI are for sure beating Anthropic on that front.

And the open weight labs are doing even better.

[Anoni Mouse @__AnoniMouse__]

@awesomekling @antirez I'm a little worried about it tbh.

If you look at our history, humans, despite having thrived as a species (so far), we frequently mess up on a grand scale too.

There are no guarantees that this AI narrative ends up a success.

[Anoni Mouse @__AnoniMouse__]

@AlexJonesax This looks cool 😎

[Anoni Mouse @__AnoniMouse__]

@djdlc_1895 @ivanfioravanti @OpenAI Pi coding agent. Job done ✅

[Anoni Mouse @__AnoniMouse__]

@ShimazuSystems I think I held out until about 30. Most people's eyesight drops off about this age.

Seeing is great! Lol
We're very lucky to have this tech tbh, easy to take for granted.

Having said this, I'm well overdue a new prescription and I keep putting it off 😂

[Pi @pidotdev]

There are many agent harnesses, but this one is open source and it is yours.

Thank you People of Pi for 70000 stars on GitHub⭐️

[Anoni Mouse @__AnoniMouse__]

@Teknium Oh dear. So sorry for your loss 🫶

[Anoni Mouse @__AnoniMouse__]

@stevibe Great work, I love your benchmarks. So creative 🙌

[Anoni Mouse @__AnoniMouse__]

@lauriewired Imo they're different levels of magnification.

Sometimes I zoom in and absorb into all of the details. On an OS like Linux you can really go as deep as you like 🤓

Sometimes I zoom out and focus on what I'm trying to build. AI is useful for prototyping and iterating quickly 🚀

[Anoni Mouse @__AnoniMouse__]

@robj3d3 @AnthropicAI Take care of yourself 🙏🫶

[Anoni Mouse @__AnoniMouse__]

@engelnyst @TimTeaFan @simonw Ok that sounds more believable. A language model is just that - there's no intelligence there - although these robots have incredible skills we need to harness.

For now I remain optimistic, in that I think we have a great future ahead of us: Humans and robots, working together.

[Anoni Mouse @__AnoniMouse__]

It will be interesting to see how things develop. Human beings have millions of years evolution baked into our neurology. It's not all desirable, but it's at present the best known general intelligence in the universe. Highly adaptable, it can even rewire itself after an injury and it runs on about 30W of power.

We'll have to see if large language models are ever able to truly generalise. We're getting great mileage atm out of reinforcement learning. Essentially taking the raw model and specialising it as an office IT worker. But we'll have to see if it's possible to get these models to thee point where they can generalise, adapt and learn.

It may end up that they are just extremely useful robots we'll all learn to control and orchestrate in our daily lives. I'm not yet at all convinced that a model of language can replace human intelligence, no matter how many parameters and CPU cores they chuck at it.

[Anoni Mouse @__AnoniMouse__]

However I'd argue this is a task, and not a job role. A human research assistant can handle a much broader range of scenarios and is able to generalise. When the online crawl hits a wall, they can pick up a phone and talk to an academic at another institution and ask for something to be sent over.

Or perhaps someone arrives at the office and says I've got a load of data here that might be useful to you, but it's all on an external SSD in PDF format.

My examples a vague/trivial because we haven't defined a domain etc - but the human researcher is able to adapt and handle a huge range of unforeseen, blocking scenarios.

By contrast, the AI process is very narrow and it will work fine within a smaller predefined scope, but it can't go off the tracks and course correct the way a human can.

[Anoni Mouse @__AnoniMouse__]

@TimTeaFan @simonw Okay, so for sure, deep research is something these models are good at and we can even write specialised agentic systems to tailor the way we want research done. I agree they are good at this.

[Anoni Mouse @__AnoniMouse__]

@TimTeaFan @simonw Could you give an example role that you think an agent like Openclaw could do? I can think of many individual tasks - but actually replacing the whole job - are you sure about that?

[Anoni Mouse @__AnoniMouse__]

@badlogicgames @pidotdev Done 🫡

[Anoni Mouse @__AnoniMouse__]

@ShimazuSystems Oh, I remember having hair, it was nice! 😅

You're looking good 🫶

[Anoni Mouse @__AnoniMouse__]

@ZackKorman Oh ok. So the report I heard about was overstating the case then :/