barrkel

19 ms·

301.
by barrkel 1y ago
It only works on Apple though it looks like.
302.
by barrkel 1y ago
There's a lot of tacit knowledge in trades that was not documented, and in many ways can't be easily documented. You can't really document how to swim or play an instrument; you need to practice and train your body. Would you
303.
by barrkel 1y ago
The heart of this is what you think tax is. If you just think it's the government stealing your stuff, then it seems very unfair to have your stuff taxed after you die. But what tax is, under another lens, is a way to divert the capaci
304.
by barrkel 1y ago
Metropolises are not soulless, as a rule. They are dynamic and exciting, with an influx of ambitious young people every year trying their best to start something. You don't need a big home space when any cafe can become your living roo
305.
by barrkel 1y ago
I had to make this decision yesterday and picked Galaxy S25. A lot lighter than my work Pixel 9.
306.
by barrkel 1y ago
And it's the New Yorker that is frequently a culprit here. Too many articles talking about the journey the reporter went on to write their article. The low signal to noise ratio is a decent chunk of the reason I unsubscribed. Too few a
307.
by barrkel 1y ago
This is in fact the goal of Russian style propaganda. You have successfully been targeted. The idea is to spread so much confusion that you just throw up your hands and say, I'm not going to try and figure out what's going on any
308.
by barrkel 1y ago
Yes, the context (along with the model weights) is the source data from which the hidden state is calculated , in an analogous way that input and CPU ticks (along with program code) is the way variables in a deterministic program get their
309.
by barrkel 1y ago
The LLM can be predict that it may lie, and when it sees tokens which are contrary to some correspondence with reality as it "understands" it, it may predict that the lie continues. It doesn't necessarily need to predict that
310.
by barrkel 1y ago
Sure, if you pay for the product, the expectation is that the data is not used for training, because that's what the contract says. And if you have a temporary chat, the data will be deleted after a day or two.
311.
by barrkel 1y ago
The recurrence comes from replaying tokens during autoregression. It's as if you have a variable in a deterministic programming language, only you have to replay the entire history of the program's computation and input to get the
312.
by barrkel 1y ago
No, that's not quite what I mean. I used the logits in another reply to point out that there is data specific to the generation process that is not available from the tokens, but there's also the network activations adding up to t
313.
by barrkel 1y ago
Sure it's state. It logically evolves stepwise per token generation. It encapsulates the LLM's understanding of the text so far so it can predict the next token. That it is merely a fixed function of other data isn't interest
314.
by barrkel 1y ago
I believe saying the LLM has a plan is a useful anthropomorphism for the fact that it does have hidden state that predicts future tokens, and this state conditions the tokens it produces earlier in the stream.
315.
by barrkel 1y ago
There is hidden state as plain as day merely in the fact that logits for token prediction exist. The selected token doesn't give you information about how probable other tokens were. That information, that state which is recalculated i
316.
by barrkel 1y ago
Hidden state in the form of the activation heads, intermediate activations and so on. Logically, in autoregression these are recalculated every time you run the sequence to predict the next token. The point is, the entire NN state isn'
317.
by barrkel 1y ago
The problem with viewing LLMs as just sequence generators, and malbehaviour as bad sequences, is that it simplifies too much. LLMs have hidden state not necessarily directly reflected in the tokens being produced and it is possible for LLMs
318.
by barrkel 1y ago
The bottleneck IME is people. It's almost never code. It's getting alignment, buy-in, everyone rowing in the same direction. Tech that powers up an individual so they can go faster can be a bit of a liability for a company, bus fa
319.
by barrkel 1y ago
It's not really true. There are many session-scoped flags you can set which influence planner behavior, and sometimes toggling them for a specific query is what gets you through. More often, judicious use of CTEs as an optimization bar
320.
by barrkel 1y ago
This isn't quite right. Even when an LLM generates meaningless tokens, its internal state continues to evolve. Each new token triggers a fresh pass through the network, with attention over the KV cache, allowing the model to refine its
321.
by barrkel 1y ago
Where do you see the major price increases coming from? The underlying inference is not super expensive. All the tricks they're pulling to make it smarter certainly multiply the price, but the price being charged almost certainly cover
322.
by barrkel 1y ago
The article talks about 50% extra, not a shortfall.
323.
by barrkel 1y ago
That voice - antithesis - is straight out of ChatGPT, like the sibling comment says.
324.
by barrkel 1y ago
I understand what the author deleted and why. I would never delete my own archive of notes, because it contains a different kind of information: howtos for things I do infrequently, current state for personal projects I rotate in and out of
325.
by barrkel 1y ago
I absolutely agree, when you look at the first order. I don't quite agree when you look at a second order, applying more compute; for example, brute forcing a combination of ideas and using a judge to evaluate them. I suspect there&#x2
326.
by barrkel 1y ago
Are they subsidizing it? Training is definitely "subsidized". Some think it's an investment, but with the pace of advancement, depreciation is high. Free users are subsidized, bit their data is grist for the training mill so
327.
by barrkel 1y ago
When you self host, are you hosting SOTA models (do you know how big or sparse they are) and are you maximizing utilization?
328.
by barrkel 1y ago
What is the lack of privacy on Claude Code? Aren't Pro and API both private and not used for training?
329.
by barrkel 1y ago
LLMs are generally deterministic. The token sampling step is usually randomized to some degree because it gets better results (creativity) and helps avoid loops, but you can turn that off (temp zero for simple samplers).
330.
by barrkel 1y ago
How much does refinements of shape, terminal velocity, target characteristics change the calculation?

More ›