pizza

Worth saying that there is a similar law for CBP to be able to conduct warrantless searches anywhere within 100 miles of the border.

The parts worth thinking about you still think about. The parts that you’ve done a million times before you delegate so you can spend better and greater effort on the parts worth thinking about.

What would you do if your CEO tells you to do something illegal? What would you do if your CEO then tells you to intimidate people who refuse to carry out the illegal requests by tailing them and then taping the surveillance footage to their door as a threat?

What do you want it to do differently? There's still the option of making a PR or a fork or sharing an alternative you've built with others.

It's a spec for how external info can get into a language model's context window. Part of that is tool calling, but there's also templated input, cancellation, progress tracking, recursive generation sampling, statefulness of tool servers, etc.

To paraphrase the Bjarne Stroustrop quote: There are only two kinds of language model tool calling protocols: the ones people complain about and the ones nobody uses

I just want to mention something in a chat in 5 seconds instead of preparing the input data, sending it to the API, parsing the output for the answer, and then doing it all again for every subsequent message.

You might be interested in the complexity metric from shape dynamics involving the ratio of intra-cluster to inter-cluster distances [0], the magnitude metric for maximal diversity [1], and haecceity [2] as repurposed for the meaning of entropy as specificity.

[0] https://arxiv.org/pdf/2201.07979#subsection.5.1.2

[1] https://case.edu/artsci/math/mwmeckes/publications.html

[2] 'thisness' - in the sense that, the entropy quantifies the bits needed to pick some named option from more options

Why couldn't talking to a brick wall be therapeutic as long as you began to open yourself up, overcoming internal resistance, etc?

I've gotten decent outputs with Claude with iteration (sending both text feedback and screenshot for context) and then tweaked the output in Inkscape.

My guess: by measuring the minimality of the amount of energy spent restoring neuron depolarization capability by way of efficient clearance of metabolic waste (and wider network adaptivity) of thalamo-cortical circuits. The hypothesis being: 'for a neuron to struggle to repolarize = thinking feels and works bad'

How to improve it? Good hydration, good vascular system architecture, good sleep, good hemodynamics, low tension, low inflammation, good glymphatic drainage during sleep, good interspersed periods of rest between thinking, good perception of one's own fatigue so you don't burn yourself further with more activity - or even better, so you know how to make yourself feel at ease and comfortable to think, etc.

https://neurosciencenews.com/thalamus-conscious-perception-2...

But isn't it actually more like, COBOL lets you talk in COBOL-ese (which is kinda stilted), whereas LLMs let you talk in LLM-ese (which gets a lot closer to actual language)? And then since the skill cap on language is basically infinite, that this becomes a question of how good you are at saying what you want - to the extent it intersects with what the LLM can do.

> On Tuesday the Commerce Department announced it was ending nearly $4 million to Princeton for programs it said exaggerated climate threats and contributed to “climate anxiety” among young people.

More like climate anxiety among energy execs; that statement is some pretty wicked doublespeak.

Claude Code + Vim (+ occasional JetBrains IDE usage) is working phenomenally for me

off topic but the timing is a doozy: tik tok to be banned again this saturday. wonder what whoever is appointed tomorrow will have to say about the dangers it poses..

Oh, I thought you were talking about unorderedness in embedding indices in a general context, to which I was responding to the specific case of vocab embedding indices having a correlation - my apologies

They do have a weak relationship, in that earlier index tokens were encountered earlier during the formation of the vocabulary, so they are similar in typicality

I mean, do Windows or macOS tell the browser which mail apps you have installed when it handles a mail:// URI?

Here’s an example of a type of feature I want: I’m looking at a menu from a popular restaurant and it has hundreds of choices. I start to feel some analysis paralysis. I say to my computer, “hey computer, I’m open to any suggestions, so long as it’s well-seasoned, spicy, salty, has some protein and fiber, easy to digest, rich in nutrients, not too dry, not too oily, pairs well with <whatever I have in my fridge>, etc..” Basically, property-oriented search queries whose answers can be verified, without having to trudge through them myself, where I don’t really care about correctness, just satisficing.

Hang on a sec... the qualia research institute might take umbrage at that first bit.

It's worth pointing out that even with the largest models out there, coherence drops fast over length. In a local home ML setup, until somebody radically improves long-term coherence, models with < x memory may be a diametrically opposed constraint to something that still says the right thing after > y minutes of search.

Checking if a recommendation system is actually good in practice is kind of tough to do without owning a whole internet media platform as well. At best, you'll get the table scraps from these corporations (in the form of toy datasets/models made available), and you still will struggle to make your dev loop productive enough without throwing similar amounts of compute that the ~FAANGs do so as to validate whether that 0.2% improvement you got really meant anything or not. Oh, and also, the nature of recommendations is that they get very stale very quickly, so be prepared to check that your method still works when you do yet another huge training run on a weekly/daily cadence.

Well, language models don't measure the state of the world - they turn your input text into a state of text dynamics, and then basically hit 'play' on a best guess of what the rest of the text from that state would contain. Part of your getting 'lies' is that you're asking questions for which the answer couldn't really be said to be contained anywhere inside the envelope/hull of some mixture of thousands of existing texts.

Like, suppose for a thought experiment, that you got ten thousand random github users, collected every documented instance of a time that they had referred to a line number of a file in any repo, and then tried to use those related answers to come up with a mean prediction for the contents of a wholly different repo. Odds are, you would get something like the LLM answer.

My opinion is that it is worth it to get a sense, through trial and error (checking answers), of when a question you have may or may not be in a blindspot of the wisdom of the crowd.

Btw, would anyone have any good resources on using pytorch as a general-purpose graph library? Like stuff beyond the assumption of nets = forward-only (acyclic) digraph

there’s a differentiable version of this that compiles to C or CUDA: difflogic

We generally don't let markets choose rights

It would probably be really great for web searching llms to let you calibrate how they should look for info by letting you do a small demonstration of how you would pick options yourself, then storing that preference feedback in your profile’s system prompt somehow.

Yup: free solar photons, and living.