jononor

35 ms·

661.
by jononor 2y ago
This can happen pretty easily with ORMs also, which is commonly used together with REST. Not that it really detracts, these risks are quite real.
662.
by jononor 2y ago
Why wasn't the original implementation swapped for the new one? The unwillingness/inability to do that seems to be most likely the core of the issues here?
663.
by jononor 2y ago
ESP32-S3 or ARM Cortex M7, probably.
664.
by jononor 2y ago
I am pretty sure a lawyer could be found that would happily take that case against the board. And that Altman would be quite willing to go that far. At least I would not have staked my professional life and personal economy on it not happen
665.
by jononor 2y ago
Minor nitpick/clarification. As it stands this is doing detection of a fixed, small vocabulary of words - not open ended text to speech covering entire language. Also called speech command recognition / keyword spotting. Which is
666.
by jononor 2y ago
Really nice project! Great care is taken in optimized audio feature extraction, very cool to see. I am working on a very similar project[1], using the Puya PY32. I opted for that chip over CH32 since it has DMA (simplifies efficient ADC inp
667.
by jononor 2y ago
It is a good idea that is worth trying out! Like anything there are tradeoffs though, so it is not guaranteed to be better for this particular circumstance. The ability to use low bitdepth integer operations (which easy for a neural net) sh
668.
by jononor 2y ago
PyTorch of course. Or alternatively a lib or custom code on top of TinyGrad. Is that a problem?
669.
by jononor 2y ago
Standardization for saddles is very high. The overwhelming majority of saddles attach to the seatpost using two metal rods of standard distance and diameter. So there is no need to suffer if your nether parts hurt after weeks of using the b
670.
by jononor 2y ago
Sure it might well be. Ads and Cloud are the main things making money at Google. Those have very high profit margins. So it might be 90% at Google are just spending the money earned by the 10% that are directly bringing in the dough :)
671.
by jononor 2y ago
It makes sense for companies to have 1-10% of resources allocated for that. At Google scale that is thousands of people.
672.
by jononor 2y ago
But current models must be retrained to incorporate new information. Or to attempt to fix undesirable behavior. So just freezing it forever does not seem feasible. And because there is no way to predict what has changed - one has to verify
673.
by jononor 2y ago
The proponents of open firmware and open hardware that I know consider built-in firmware as bad-but-slightly-better-than-updatable. Slight better because it is theoretically easier to validate, as one does not need to take into account firm
674.
by jononor 2y ago
Encyclopedia Britannica is also wrong in a reproducible and fixable way. And the input queries a finite set. It's output does not change due to random or arbitrary things. It is actually possible to verify. LLMs so far seem to be enti
675.
by jononor 2y ago
Exactly. And if that is managed, perhaps the codebase can be further adapted on the future to actually open hardware.
676.
by jononor 2y ago
Not great by modern standards... The standard ESP32 supports 802.11n, but only 2.4 GHz (not 5). So theoretical rates maybe up to 40-50 Mbps. In tests that I have seen people do on ESP32 they have reported practical speeds of 5-10 Mbps. Whic
677.
by jononor 2y ago
Very cool project. I think efforts like these are among the most promising to get a FOSS WiFi cards. Because these WiFi-capable MCUs are: designed to be generally programmable, have quite a lot of open documentation, are available for purch
678.
by jononor 2y ago
It is getting more and more like the Matrix. Just that we are being harvested for our data (behavioural patterns), not the energy.
679.
by jononor 2y ago
You can use Linux to have a desktop/laptop PC that respects you r privacy. Btw, this shit is likely to go into Android as well. Maybe iOS will stay clear for a bit longer...
680.
by jononor 2y ago
A few kilometers of snow is easily crossed by skis or snowshoes. Every cabin in Norway used to be accessed like that (changing now the last 2 years, where people now mostly expect to drive to the door). But still skiing a few kilometers is
681.
by jononor 2y ago
It some areas superhuman performance is sadly not that high of a bar...
682.
by jononor 2y ago
It is worth noting that what is doable/practical with entry level CAD has increased a lot over the last 20 years. Hobbyists will not feel restricted, and one can do quite a lot of commercial relevant work also.
683.
by jononor 2y ago
Assuming 10 Mbit/s per stream that would serve 1 million concurrent streams 24/7. If we assume people watch 2.4 hours per day on average, it could support 10 million active users. Or a cost to serve for bandwidth alone of 1 USD&#x
684.
by jononor 2y ago
Depends on what you mean by learn... And what other relevant background one has outside of CS - programming/math/stats/systems/engineering. Assuming one has a few years of that, and one want to get to the level of being
685.
by jononor 2y ago
Great tip! Making Rochelle's salt definitely looks very doable. I found a video on YouTube that grew crystal and used it as mic on a guitar. Sounds pretty good to be honest! Though the form factor is a bit odd - a bulky crystal blob :D
686.
by jononor 2y ago
Not so may people want "AI art" (branded as such). But a good chunk people use generative image model to "commission" works that they are interested in - but would be to expensive to get a human artist to do it. There ar
687.
by jononor 2y ago
Do you not consider such collectives part of we (the humans)? I can understand arguing that organizations have some superinteligence. But they are still made of / run on humans - not AI/computers?
688.
by jononor 2y ago
This is some internal reference/joke. We on the internet don't really know what it actually means.
689.
by jononor 2y ago
Anyone or anything at all is probably the first steps, which we still don't know how to do - even conceptually. But the first real target would be alignment with the creator of the AI, and the goals an values they set. That is: Alignme
690.
by jononor 2y ago
Many teams are trying to combine their ideas with LLM. Because despite their weaknesses, it seems LLMs (and related concepts such as RLFH, transformers, self-supervised learning, internet-scale datasets), have made some remarkable gains. T

More ›