sdenton4

10 ms·

181.
by sdenton4 10mo ago
Sure, but extremely cheap collared with the us. Here's a UNICEF report comparing childcare policy amongst rich countries: https://www.unicef.org/innocenti/reports/where-do-rich-count... Sweden ranks third in
182.
by sdenton4 10mo ago
I believe Sweden is one: https://www.sdg16.plus/policies/universal-childcare-model-sw...
183.
by sdenton4 10mo ago
Mike Waltz is just about dumb enough to pile out his own eyes with his thumbs. At which point we will be regaled with the danger of thumbs forevermore.
184.
by sdenton4 10mo ago
Signal is already complication free (at least until your phone falls in a lake) making the control useless. (And you probably don't need to worry about losing the 'running late' message in the lake... The need for good encryp
185.
by sdenton4 10mo ago
This mod includes collecting cheese (though it's a side activity, and maybe too subtle for the rats): https://youtu.be/qPRvw6kRN-8?si=j9iuTuiHerm0AhQ2 And here's a thing I knew had to exist: a doom mod/level
186.
by sdenton4 10mo ago
Feels like the "right" approach would be to have it write some code to measure how far off the elements are in the original vs recreated image, and then iterate using the numerical output of the program...
187.
by sdenton4 11mo ago
* Targeted tariffs and blanket tariffs are different beasts. * In order for capitalism to undercut the tariffs, the tariffs need to be high enough to offset the costs of setting up the local industry and the higher costs of US labor (which,
188.
by sdenton4 11mo ago
AI had a kind of Jevons paradox approach to efficiency improvements, unfortunately - if you halve the compute requirements with an algorithmic advance, you can run a model twice as big.
189.
by sdenton4 11mo ago
Those independent companies can be eventually bought by PE and turned into more bombs, though, right?
190.
by sdenton4 11mo ago
I think this is a bit overblown. In the event of a crash, the current generation of cards will still be just fine for a wide variety of ai/ml tasks. The main problem is that we'll have more than we know what to do with if someone
191.
by sdenton4 11mo ago
i've got laser eyes and i know what you're thinking comes as no surprise that christmas lights are blinking https://www.youtube.com/watch?v=AJWSu9M4ffQ
192.
by sdenton4 11mo ago
It strikes me that neutral network inference loads are probably pretty resilient to these kinds of problems (as we see the bits per activation steadily decreasing), and where they aren't, you can add them as augmentations at training t
193.
by sdenton4 11mo ago
Microorganism A is good at producing (say) insulin, but needs oxygen to do it. Microorganism B is good at ferrying oxygen around. Microorganism C is good at extracting oxygen from the air and attaching it to B. Each of these cells has speci
194.
by sdenton4 11mo ago
You can factor humans out with the same trick: All economic activity is driven by the needs of micro-organisms. We mainly observe these desires and economic transactions through their conglomeration into joint entities (ie, humans). Corpora
195.
by sdenton4 11mo ago
Meanwhile, corporations, ngos, and governments are not exactly people, and partake in economic transactions all the time.
196.
by sdenton4 11mo ago
Yep! Think of LORA for network fine tuning. Monarch (linked above) uses lots of block diagonality. These ideas also make flash attention flash. I haven't seen banded matrices as much, though (with weight sharing) they're just con
197.
by sdenton4 11mo ago
Heat is literally inefficient by definition - it's spent energy which hasn't been converted to visible light.
198.
by sdenton4 11mo ago
A somewhat more beautiful matmul for neural networks is given by the Monarch paper: https://arxiv.org/abs/2204.00595 Generally, low-rank and block-diagonal matrices are both great strategies for producing expressive ma
199.
by sdenton4 11mo ago
The register might already do the rounding if it was designed to work in Canada, which got rid of the penny over a decade ago.
200.
by sdenton4 11mo ago
We can turn them into suntan lotion!
201.
by sdenton4 11mo ago
Maintaining safety across multiple driving "modes" multiplies the complexity of the problem, in a space where safety incidents can shift public opinion of the whole industry. This is a bad idea...
202.
by sdenton4 1y ago
s/LLM/human/
203.
by sdenton4 1y ago
The problem is that serving content costs money. Llm scraping is essentially ddos'ing content meant for human consumption. Ddos'ing sucks.
204.
by sdenton4 1y ago
Why is that the baseline? Actual human performance as it exists today gives us tens of thousands of road fatalities per year in the US. We have not solved that problem despite decades of opportunity to introduce regulations and enforcement.
205.
by sdenton4 1y ago
Hm, so you would put a hypothetical scenario on the same footing as thousands of actual deaths caused by drunk drivers each year? 30% of us road fatalities involve a drunk driver each year... I seriously doubt that the "mass takeover a
206.
by sdenton4 1y ago
Why should we exclude the cases of human bad-actors? That's explicitly a major case solved by getting rid of the human behind the wheel...
207.
by sdenton4 1y ago
/I'd love to see new approaches that explicitly don't "admit a GPU-friendly formulation", but still move the SOTA forward. Has anyone seen anything even getting close, anywhere?/ The speedup from using a GPU ov
208.
by sdenton4 1y ago
Eventually someone will make a killing on doing actual outcome measurements instead of just trusting the LLMs, Michael Lewis will write a popular book about it, and the cycle will begin anew...
209.
by sdenton4 1y ago
And yet, people very often find themselves using generative models for categorization and information retrieval tasks...
210.
by sdenton4 1y ago
I see a fair amount of bullshit in the LLM space though, where even cursory consideration would connect the methods back to well-known principles in ML (and even statistics!) to measure model quality and progress. There's a lot of &#x2

More ›