barbegal

13 ms·

1.
by barbegal 4d ago
Without a human to benchmark against it's really tough to gauge how good these models are vs how good the task definitions and existing codebases are. My intuition from the example full instructions are that the tasks are poorly specif
2.
by barbegal 2mo ago
18 O enriched water is available to buy but the pricing suggests it's quite expensive to separate 16 O from 18 O. And you get much less variability in the ratio of isotopes from mixing compared to separating a single isotope as much as
3.
by barbegal 4mo ago
Interesting but I can't deal with the AI generated prose. Sentences like "This was not a garment for walking between a carriage and a doorway. It was ..." show me no human has proof read or edited the text.
4.
by barbegal 5mo ago
Yeah I couldn't get past so many issues in the AI generated illustrations. Not useful at all when they are completely wrong.
5.
by barbegal 5mo ago
Yes I would imagine lots of those type of services would be vulnerable if they hadn't updated to the latest kernel versions.
6.
by barbegal 5mo ago
Yes you can pin it to core 1 whilst pinning all other tasks to core 0. Then will never be interrupted or preempted (except by interrupts created on core 1)
7.
by barbegal 5mo ago
The actual research paper shows it's pure nonsense https://cdn.skoda-storyboard.com/2026/04/Skoda-DuoBell-Resea... As expected ANC headphones cancel less noise at low frequencies so I guess the 780Hz is a tra
8.
by barbegal 6mo ago
I don't understand why people get so hung up on Chrome using so much memory. A lot of this memory is "discardable" so will get dropped when the system is under memory pressure and the amount of memory allocated for this type
9.
by barbegal 6mo ago
Does the KV cache really grow to use more memory than the model weights? The reduction in overall RAM relies on the KV cache being a substantial proportion of the memory usage but with very large models I can't see how that holds true.
10.
by barbegal 6mo ago
This series of graphs https://www.bmj.com/content/bmj/387/bmj-2024-082194/F1.large... shows that whilst those two professions are at the bottom of the distribution they are not particularly outlying and
11.
by barbegal 6mo ago
Total receipts were over 11,000 so more like 100 hours or around $2000 so a similar price to the LLM.
12.
by barbegal 6mo ago
This is good work. I wish branch predictor were better reverse engineered so CPU simulation could be improved. It would be much better to be able to accurately predict how software will work on other processors in software simulation rather
13.
by barbegal 7mo ago
The memory overhead is fairly significant it uses between 1.5 and 3 times the space of the data stored.
14.
by barbegal 8mo ago
For a real spinning top over engineered https://youtu.be/QLTsxXNekVE?si=S31kpZQHiYlUSedx
15.
by barbegal 8mo ago
Only 2 students actually used an LLM in his exam, one well and one poorly so I'm not sure there is much you can draw from this experience. In my experience LLMs can significantly speed up the process of solving exam questions. They can
16.
by barbegal 8mo ago
An interesting idea but in theory just three correct pass codes and some brute force will reveal the secret key so you'd have to be very careful about only inputting the pass code to sites that you trust well. It's definitely comp
17.
by barbegal 8mo ago
The energy comes from the metal electrodes not the potato. Potato is just an electrolyte carrying current between the cathode and anode.
18.
by barbegal 8mo ago
https://newsletterhunt.com/emails/215242 seems to be a copy
19.
by barbegal 9mo ago
Congratulations! Works amazingly well. Does it generate enough revenue to be self sustaining?
20.
by barbegal 11mo ago
That 90,000 traces did take 225 hours to capture so it is truly a huge amount of data and not a trivial attack.
21.
by barbegal 11mo ago
Thanks, good to know that I'm wrong!
22.
by barbegal 11mo ago
Is it just me or does this have a familiar "edited by Chat GPT" feel to it? I can only take this chatty writing style in moderation but it seems everyone is using Chat GPT to edit their work in the same way.
23.
by barbegal 1y ago
Yeah as far as I can see all the companies that study EV batteries and provide degradation reports etc. all do so by using the data from the manufacturer. I would trust data about battery degradation a lot more if the data came from an inde
24.
by barbegal 1y ago
The ForTwo has two seats and no truck bed so it can have enough space for a crash structure in front of the driver. On top of that its NCAP rating expired in 2021 and it is no longer being sold. The next generation of the ForTwo is likely t
25.
by barbegal 1y ago
There are smaller vehicles being sold but the distance from the front bumper to the driver's legs is much longer because they don't have a bed taking up space at the back.
26.
by barbegal 1y ago
The reason why we don't generally have vehicles this small any more is because they don't pass crash tests so I'm wondering how this fares in a crash test. I can't see any way this could be sold in Europe unless there&#x
27.
by barbegal 1y ago
Assuming you could get these to 10% efficiency (which is theoretically possible) and a phone needs 0.2W of energy to function then you would need a source capable of supplying 2W of energy (of which 1.8W would be dissipated as heat). The ph
28.
by barbegal 1y ago
Not comparable to a real battery any time soon based on the paper https://pubs.rsc.org/en/content/articlelanding/2025/cc/d4cc0... > The perovskite betavoltaic cell achieved impressive parameters,
29.
by barbegal 1y ago
Only 1 of the four turbines has been able to operate for 6 years without pulling it out the water. The other 3 have needed costly maintenance https://www.waterpowermagazine.com/news/sae-secures-loan-for... It's a
30.
by barbegal 1y ago
In each loop prepending a single character could take O(m) (moving all m characters one to the right) so combined O(nm) where n is the number of padding characters and m is the total number of characters in the string.

More ›