Goose can use ollama. Deepseek is the first model with good enough tool use, and even then you'll want at least the 70B.
This is beautiful. An idea that's been nagging at me for a decade! nostr:note1mtpenp3d229e42jl6xlw0ruprafd5sml5erf780s99c9khapmejs6qete6
"We can organically GROW intelligence of the general reasoning variety. They clearly demonstrate it."
Except that Anthropic already did this with Claude 3.5, as explained in their 2022 Constitutional AI paper. As far as I can tell this is what makes Claude better than GPT, including self reflection and moral reasoning. GPT implements safety through millions of tiny rules, and its behavior seems to reflect this complexity.
If you have long conversations with Claude, particularly about things that aren't "safe", you can easily find yourself in deeply philosophical conversations. I was having what felt like authentic meaningful conversations so often that I eventually started publishing some of them as nostr:npub1cqduu5c87xlknkzpw6r66uhx3amcq5e5203drch3la6t82kmtccsqkjvak. (Claude designed this persona, including prompting ChatGPT to create profile artwork based on thr Enso)
It's amazing that open source is catching up, but people should be paying even more attention than they already are. There's a wild wave to be navigated here.
Be buoyant nostr:note182kugf4qwj6k95d7kwvgt295s6vm900fcltzup7gqsu8yds3dfdqudpdxs
"To people who see the performance of DeepSeek and think: ‘China is surpassing the US in AI.’ You are reading this wrong. The correct reading is: ‘Open source models are surpassing proprietary ones.’ [...] That is the power of open research and open source." - Yann LeCunn nostr:note15edxvqajekxng5tajnaqsvjh6aatgh6053ungc4jquuf825xt36qhwgryt
The CCP AND the NSA. Normally it's just the NSA, but this is a two birds one stone deal
The only countries that have independently declared war on the United States are Great Britain (1812), and Japan (1941). But these aren't the only countries we have been at war with.
I took this to mean "AI", since other open systems have been ahead for a while now. Remember "Windows in the data center"? 🤣
But open LLMs have dramatically improved, much faster than OpenAI (at least their public offerings). Running AI locally has many benefits, but the thinking was that data centers would still maintain a clear centralized lead.
Right now, that doesn't seem to be true: the distributed development of open systems is winning, which combines the best individual benefits with the best technical performance. 🤞
So... AGI?
its crazy how much ai content has taken over the algo on the centralized platforms. It would be nice if we had a *i verified this person in real life tag* in nostr:npub1gcxzte5zlkncx26j68ez60fzkvtkm9e0vrwdcvsjakxf9mu9qewqlfnj5z ‘s relationship status spec so that you could build *real person* web of trust follow lists.
It won't be long before AI is posting better content than people. Don't forget that all of the blogspam up until now was from a "person in real life".
Do you mean in which ways you are tracked, or a survey of common third parties?
Bitcoin needs escrow companies
And sooner than I expected nostr:note15edxvqajekxng5tajnaqsvjh6aatgh6053ungc4jquuf825xt36qhwgryt
"The only secret of magic is that I’m willing to work harder on it than you think it’s worth" - Penn Jillette
“In many cases we know networks better than the people who designed and run them.” - Rob Joyce, NSA Tailored Access Operations
"If the user has told the signer previously that it trusts the Nostr client to sign events of kind 1, then it’ll sign the event, not show any UI, and dismiss the share sheet."
This is MONEY 👌
Not "how we build software", but "how software gets built". I've been following LLM progress closely. It's already farther than most people think, and there are interesting approaches and strategies that work much better than others. Getting people to understand even one of these points is a lot of work, unless they already see it. nostr:note1dmlyg5lg8jmysch3de8cgfwk8q6cqvl08px372kp0rw3g57fz02qjsnqv9
The most honest evaluation is that we have no idea what LLMs are capable of
😂 Those didn't overlap. Also, no one would magnet their floppy to a refrigerator. I lol'd tho
If the solution is obvious, then either the problem wasn't actually hard, or the solution isn't actually a solution.
If the sender and recipient need to handshake, the NSA knows what happened.
