frabcus

10 ms·

121.
by frabcus 1y ago
Act coherently in an agentic way for a long time, and as a result be able to carry out more complex tasks. Even if it is similar to today's tech, and doesn't have permanent memory or consciousness or identity, humans using it will
122.
by frabcus 1y ago
Firstly it can render environments in detail. I'm (mostly) aphantasic even in dreams, so this wasn't obvious to me. But most people literally get visual renderings in their mind. Secondly, it's fairly clear now that our sen
123.
Big Carl lifts the dome onto second reactor – Hinkley Point C [video](youtube.com)
1 points by frabcus 1y ago|0 comments
124.
by frabcus 1y ago
They have somewhat an internal model of arithmetic, with lookup tables and separate treatment of digits. I'm conscious you might have seen this already and not interpret it like that, but in case you haven't section 6 on addition
125.
by frabcus 1y ago
Reflecting on this whole situation, I suspect MCP is fundamentally insecure, in which case Supabase should refuse to implement it. MCP's goal is to make it easy for end user developers to impulsively wire agentically running LLM chat
126.
by frabcus 1y ago
I think it's a flaw in end-user MCP combined with agentic, where the end-user chooses the combination of tools. Even if the end-user is in an IDE. The trouble is you can want an MCP server for one reason, flip it on, and a combination
127.
by frabcus 1y ago
Anthropic call this "workflow" style LLM coding rather than "agentic" - as in this blog post (which pretends it is about agents for hype, but actually the most valuable part of it is about workflows). https://
128.
by frabcus 1y ago
For the record, I find the witty loading phrases a real turn-off. It makes me feel like I don't know what it is doing, and it is being patronising.
129.
by frabcus 1y ago
Is there any way we can get local tokenizers for other LLMs? e.g. Gemini only offer a remote API for their tokenizer. Is it proprietary? Could we infer the token mapping somehow efficiently by making lots of calls?
130.
by frabcus 1y ago
It is different, because productive output does not go down by 20% when working hours go down by 20%. e.g. See "Four-day week trial: study finds lower stress but no cut in output" https://news.ycombinator.com/item?
131.
by frabcus 1y ago
Presumably you can use them all in a browser?
132.
by frabcus 1y ago
Block the adverts, and pay the creators via Patreon. And join Nebula to build other alternatives.
133.
by frabcus 1y ago
Looks like the assumption is $100 / year for each user, which with 50k users makes the $5 million ARR. Then you have to pick a finger in the air multiplier for the value of the business. A stockmarket listed SaaS company that isn'
134.
by frabcus 1y ago
Because it isn't disabling them - that would be an option to actually disable them. Of course there are hacks round it, but ultimately Google want us to have the AI summaries, don't care about user choice, and the best course of a
135.
by frabcus 1y ago
Absolutly you can - the UX of the whole experience MCP is part of could make it clear to the user what repositories can be accessed according to the project they're working on. Rather than when they're working on the public projec
136.
by frabcus 1y ago
It's a fundamental user experience flaw with the MCP server. It does indeed need fixing - e.g. it could have a permissions system itself, so even if the GitHub token has permissions, different projects have their tool calls filtered to
137.
by frabcus 1y ago
This somewhat happens already, with system messages vs assistant vs user. Ultimately though, it doesn't and can't work securely. Fundamentally, there are so many latent space options, it is possible to push it into a strange area
138.
by frabcus 1y ago
I'd like version numbers to indicate some element of backwards compatibility. So point releases (mostly) wouldn't need prompt changes, whereas a major version upgrade might require significant prompt changes in my application. Thi
139.
by frabcus 1y ago
The option that strikes me as missing, is making users pay a cost before they are randomly entered in a lottery for the ticket. So, for example, everyone pays $0.01 on their credit card, or does a holding charge on their credit card, or reg
140.
by frabcus 1y ago
"My top priority today is to internalise data structure B"
141.
by frabcus 1y ago
Even then the human doesn't know how they execute the algorithm, or mix the colours together - our conscious self-reflective mind has limits as to how far into our neural network weights it can reach. Can get further with lots of med
142.
by frabcus 1y ago
They mention Google Cloud Marketplace (not Google Cloud Platform), this seems to be their listing there: https://console.cloud.google.com/marketplace/product/mistral... Which says: "Managed Services are fully
143.
by frabcus 1y ago
There's an in between case, where LLMs are useful and give coders a (say) 20% speedup, and everyone has to use them. They don't have to be perfect to be a big industry!
144.
by frabcus 1y ago
Where's this non-compete clause? In ChatGPT T&Cs?
145.
by frabcus 1y ago
Yep - I switched from iPhone to Android and Tasker is much more powerful. Slightly less good UX and mainly systems focussed. But actually allows any trigger. Also extensible - it can trigger scripts in Termux. You can install any command li
146.
by frabcus 1y ago
Presuabmly the browsers will still get paid by other search engines to set default? This could ultimately help DuckDuckGo, Bing and Kagi quite a lot. And the revenue won't fall completely.
147.
by frabcus 1y ago
Thunderbird is (amazingly) funded entirely by donations from users. The money is hypothecated for that purpose. There are so few good email clients now they get enough donors. Source: someone I know who works there
148.
by frabcus 1y ago
Right but presumably 85% of the parts aren't imported? So while it is a benefit, it is a slightly bizarre one? Would be nice to see a technical definition for how the % imported is worked out.
149.
by frabcus 1y ago
You don't have to consciously and rationally understand the grammar - you didn't when first learning to speak your first language! Stephen Krashen is a pretty good researcher on this - the summary is that exposure to the language
150.
by frabcus 1y ago
This new license makes it completely clear that LLama models are, and never have been, open source. They are "shared source" or "open weights". There's no freedom with these models.

More ›