nok22kon

or the cache retrieval function for a key retrieving the wrong entry

I found LLMs very good at reading human chat logs and doing psychological analysis of the participants, and of the interplay

hard to imagine how they do that without a good understanding of emotions/theory of mind

if they pattern match to psychology books, that is still functional understanding since they can operationalize them

Anthropic is selling API tokens at 80% margin.

And API is 80% of their business (subscriptions the other 20%)

you assume that the error will always be in one direction

and if sometimes you ventilate a bit sooner than required, at 700, what?

businesses will not put $200 meters in every room

could be a selection effect at work

how can you detect without a study if CO2 meters are basically nowhere?

https://pmc.ncbi.nlm.nih.gov/articles/PMC4892924/

if you just want to know if CO2 is too much, 300ppm precision is fine.

I dont need to know the exact level, just give me a green/yellow/red LED and make it cheap so I can have a sensor in every room

big AI labs make so much money because they have a good (amazing) product

curious if you tried local LLM models for OCR, like a Gemma4, or your volume is too much for that

> we are able to make the rock calculate

I'm so tired of this cliche...

a CPU is a rock about as much as much as an airplane is a rock (aluminium ore), "look, we are able to make rocks fly on their own power, isn't that awesome?"

people using it think they are impressing the normies, but it just shows them being condescending "of course I dont believe rocks calculate, but I assume you are so stupid and know so little that you might actually believe me and be impressed by my people"

plot twist: the vast majority of "studies" are like this, complete garbage, but called "science" because they put a table with p-values in

damn, good observation, when my data analysis python script goes wrong and allocates 24 GB of RAM on a 32 GB computer, it crashes (gets killed) with "out of memory" error. I've never seen something else getting killed

basic supply/demand

demand moves to cheaper providers, they ran out of capacity and increase price

literally why "commodities" are "commodities" - the more interchangeable a product is the more exposed it is to supply/demand mechanics

you will not want 20 different accounts at 20 providers, you'll have one at a "token arbitrage" provider - OpenRouter

drop-shipping is a form of arbitrage

where do nuclear weapons fit in? do they make markets more/less efficient/competitive?

next stage - a single AI which computes the "correct" price that everyone else agrees on, and instantly reprice all financial instruments without trading them, thus not paying transaction costs

assuming the typical "spherical market in a vacuum where the agents are maximally rational non-biological utility maximizers"

or maybe compute allows simulating a lot of possible cooperation strategies, and arriving at the one maximizing profits for the colluding parties

so what is a software factory then?

anthropic has 80% margin on API tokens

the more portable everyone is, the more arbitrage will happen - cheap providers will raise prices too because users will flow to them

how exactly did you disabled it on Windows?

I dont think it has an option for that.

its one thing to say "we dont know"

it's a different thing to say "it's mathematically impossible"

so if it turns out it is possible, what then? was math broken? or the researcher an idiot who either doesnt know math, or is just bullshitting non-existing proofs?

this uses 2 ADCs:

https://tascam.jp/int/feature/32-bit_float

you can also freely download CP on your private property PC

but I wouldn't advise it

and CPUs have the same basic architecture since 2000. no progress happened, right?

Yann LeCun was saying 3 years ago that because token generation is auto-regressive, its mathematically impossible to generate a long stream of coherent tokens, because errors amplify exponentially.

and then models learned that they can back track and error correct

so much for "mathematically impossible..."

I have seen it.

This is a marketing page after all, explaining point-wise samples, reconstruction filters and "staircases" is way beyond scope.

Same as in images, pixels are not "little squares"

> In a 32-bit float recorder, you have two ADCs working in tandem to create a single audio file. One “low gain” ADC is optimized for high-level audio, and the other “high gain” ADC is optimized for low-level audio. If the high gain ADC clips due to loud sounds, the low gain ADC does not. And if sounds are too quiet for the low gain ADC to capture clearly above its noise floor, the high gain ADC still has plenty of headroom above its noise floor. Said another way, the low-level ADC handles the quieter sections, and the high-level ADC handles loud sections.

https://tascam.jp/int/feature/32-bit_float

you have 22 bits for the typical audio voltage level, which you call 1.0 float

but what if you "allow" double that voltage and call it 2.0 float? a strong pressure into the microphone generates a stronger voltage

thermal noise limits you on the quiet signals, but not on the powerfull ones

so 22 bit for typical -1.0 -> 1.0 range and you can add a few more bits on top of that for stronger audio pressures (voltages) which you would traditionally clip