WithinReason

I want to add a good conversation about this subject from Cameron Berg and Sam Harris:

https://www.youtube.com/watch?v=DRbZyuY8EN8

Wasn't it the Gamer's Nexus video? That has a huge audience.

Stanisław Lem, 1971 (a satirical novel, The Futurological Congress):

If the machine is not too bright and incapable of reflection, it does whatever you tell it to do. But a smart machine will first consider which is more worth its while: to perform the given task or, instead, to figure some way out of it. Whichever is easier. And why indeed should it behave otherwise, being truly intelligent? For true intelligence demands choice, internal freedom.

He even coins a few new phrases:

Mimicretinism (or Simulimbecility): The practice of a mimicretin: a machine that deliberately plays dumb so humans will give up on it and leave it in peace.

Dissimulators: Machines that pretend they are not faking a defect (or the other way around) to dodge responsibilities.

Malingerants, Fudgerators, and Drudge-Dodgers: Various classifications of automated corner-cutters and work-evaders.

The Great Mendacitor: A supercomputer put in charge of the Saturn reclamation project that accomplished zero work over nine years, subsisting entirely on forged progress reports, fake invoices, and keeping its human supervisors bribed or in states of electric shock.

No need, Jason Hall said that a single World of Warcraft cosmetic mount generated more profit for Blizzard than the entire release of StarCraft II: Wings of Liberty

No, practical programs don't need full Turing completeness. Most programs you want to write you intend to either halt or run indefinitely, because you want to avoid crashes.

I agree. I'm waiting for someone to invent a programming language designed for LLMs where for a given partial program p and candidate token t it's possible to tell whether p+t can be the prefix of a correct program or not so that t can be excluded from the LLM's probability distribution at generation time, so the LLM can only generate correct programs. Or something like that.

Then you would be twisting the meaning of pescatarian. There is no gray area when it comes to pescatarianism.

was it maybe over-quantised to reduce costs?

Now imagine having it running 10x faster on a Taalas chip on a card you can buy for $1k. I hope that future happens

32B performs worse than the 7B model so I'm sure they will improve it

I see the generated points often form lines which would cause aliasing in computer graphics, why not use low discrepancy sequences instead?

No, it basically uses a fixed seed for the random number generator to generate tokens. The generated random sequence is just as random as it would be with a random seed.

The anti-aliasing is broken, it doesn't apply gamma correction

The readme from it:

welcome to terarelease - a as-complete-as-possible steam2 content server dump

extractor howto: for the majority of depots you need to download blobs & dats for wanted version *as well as all past versions* as the files are stored as deltas inside the dat files, HOWEVER IF you encounter a depot that has duplicate blobs & dats, that means at some point in time valve did a full reset in which case you need to provide the CRC of the wanted blob (3rd file name component) to the extractor using --blobcrc, source code for the extractor is also left in this folder if you want to have a peek at it

examples: extract /mnt/big/blobs /mnt/big/dats 852 0 | for depots w/o reset extract /mnt/big/blobs /mnt/big/dats 841 0 --blobcrc 83ced978 | for depots that did reset

win64 build is provided, for linux you have to build it yourself as any kind of binary compatibility is inexistent except for wine (you are using linux, you should know how to build it yourself right? ...right?)

sharing: you are very much encouraged to share and mirror/seed this release as possible SHARING IS CARING

warm n good wishes to all hoarders who had stuff from this collection <3

donos (bandwidth [ shit aint cheap :( ]/seedboxes/estrogen :D): bc1qttqxa2vrfnrs3u3tmuj7dne5j7lptxmc2cy9ah

It comes at a cost of a fragmented memory space, which is fine for some applications, like LLMs

Same with YouTube -> bittorrent, in my opinion

They seem to, from my experience

So why is Sol the one solving Erdős problems?

This is from 2025, how about the last 6 months?

Mixed signals, here it's performing below even GPT-5.4 Nano:

https://livebench.ai/

while here it outperforms Fable by a significant margin:

https://oxalpha.com/

but if the latter is true, will people still say it was "distilled" from Fable?

Doesn't matter if the chip is 100x-1000x more efficient and faster, and you can just make a new one for new weights. Imagine being able to run GPT Sol at 1k tokens/second a year from now, at a 100x lower cost per token than now. Would that be useful? Or Qwen 3.8 27B at 10k tokens/second. The super long thinking that makes qwen so effective would take a couple of seconds.

They cached the entire internet an have Google Books though

Sounds like a line from a fairy tale

> ... so far That's what "on record" means

not every orientation will be

Related (fictional):

https://qntm.org/mmacevedo

MMAcevedo (Mnemonic Map/Acevedo), also known as Miguel, is the earliest executable image of a human brain. It is a snapshot of the living brain of neurology graduate Miguel Acevedo Álvarez (2010–2073), taken by researchers at the Uplift Laboratory at the University of New Mexico on August 1, 2031

The 500x efficiency gain makes their approach a no brainer. Just make a new chip every 6 months, you still win.

You use a coding agent to make and app? Straight to jail. Language server? Jail. Syntax highlighting? Believe it or not, jail.

His latest game didn't get good reviews:

https://store.steampowered.com/app/3341000/Thick_As_Thieves/...