Job alert: @hyphacoop is hiring a Business Development Lead, on contract for April-June 2024 for our sister project Distributed Press!
https://hypha.coop/openings/business-development-lead/ #cooperatives #jobs #dweb
Me, an idiot: “So, kids, by setting the thermostat a little lower and eating less meat, we’re doing our part to make the world more sustainable”
VCs, very smart: “We just raised $100 billion dollars from the sovereign wealth funds of three petrostates to build the world’s largest AI supercomputer. It uses as much power and water as Guatemala and the primary use case is for management consultants to autogenerate powerpoints for justifying mass layoffs.”
Video of my talk about making an #OpenSource #LLM perform function calling on my machine.
@fleeky Oh yeah! My mind goblin demo uses JavaScript so you could use any javascript function that takes json and outputs json
@fleeky It was a freeform demo going through my code so there's no slides, but I think it was recorded so I'll ping you when I get a copy
@andia I didn't ask *that many* people so maybe there was more but also maybe AI people in Ottawa just aren't into alternative social media 🤷
@fleeky I actually used Vosk for my talk today with my wikipedia enabled agent and it's way faster and more accurate than whisper (for the small models)
@fleeky Stuff like searching github, running shell scripts, checking my emails. Generally individual function calls that take in and output text
@fleeky the code they generate is also often subtly or glaringly wrong
@fleeky Not yet! It can generate text based on some initial text. You can tell it to generate somw code. You can also tell it that if it generates some JSON in a specific format that you'll call the function for it and put the result back in for it to generate more text after. It could be possible for it to call a function thar saves some code and another function to invoke it, but it's harder to keep it focused long enough with small models
Seeing some echos today of the AI trust crisis I wrote about back in December: it's very, very hard to convince people that their private data isn't being used to train AI models once they've decided that it might be happening
https://simonwillison.net/2023/Dec/14/ai-trust-crisis/
@fleeky Unsure what that would entail. :P
Gonna be presenting a demo of teaching local a #llm to search wikipedia with "Function Calling"
We have been trying to get the council to act on the smell from the chip shop near us for years. The one and only council official who deals with this stuff literally lives an hour away in the country.
He came tonight and first tried to say the smell wasn’t that offensive. Like “it is just chargrilling”. And we were like “no it’s grease”. Then I started reading from this webpage about “statutory nuisance” and arguing how this constituted one, and he was basically shamed into agreeing and promising to take action https://www.gov.uk/guidance/nuisance-smells-how-councils-deal-with-complaints
Incredible to me that in the 21C we still only rely on the subjective judgement of a human nose. There is no objective criteria for dealing with this kind of pollution.
@mortendk @eleventy I convinced my work to use eleventy for their site. Moved off of Drupal as I knew I could make a better faster site if I didn’t have to wrestle with Drupal templates. It’s a professional hospital site. Uses TinaCMS for editing content. https://fetus.ucsf.edu/
I'm gonna be in #montreal next week from monday to wednesday if anyone wants to do a lil social meet or something for dinner or lunch :)
Occult Enby that's making local-first software with peer to peer protocols, mesh networks, and the web.
Exploring what a local-first cyberspace might look like in my spare time.