loudmax

21 ms·

241.
by loudmax 2y ago
I like the term open-weights. The "source", analogous to the source code for a program, should include the training data. In this case that isn't open. The resulting weights are open, insofar as they can be redistributed an
242.
by loudmax 2y ago
I think you're right that demand for inference is likely to grow by orders of magnitude in the medium term, but I don't know that Nvidia's lead over their competitors is likely to remain as strong as it is today. That is, th
243.
by loudmax 2y ago
From 90s into the video: "Particularly our clothing, which accounts for up to 10% of carbon emissions." That claim seems kind of outlandish, and it doesn't jive with other data I've seen. If you want to buy Patagonia, t
244.
by loudmax 2y ago
Depends on the logistics. If I were confident about the security, I wouldn't mind letting my GPU participate in a distributed effort to significantly improve an open source model. This should be a few dollars a month on my power bill
245.
by loudmax 2y ago
> Let's face it, though: these models are not more dangerous than Wikipedia or the Internet. These models are not custodians of ancient knowledge on how to cook Meth. This information is public knowledge. I don't think this is
246.
by loudmax 2y ago
A lot of us nerds here on HN would probably struggle to detect sarcasm 75% of the time.
247.
by loudmax 2y ago
I think it's an interesting distinction, but I don't think everyone is one or the other. You can be both types, and (sadly) some people are neither. You can be a lyrics person when you listen to Bob Dylan, and be a music person w
248.
by loudmax 2y ago
Getting my Nest thermostat to work with Home Assistant was a lot of work. I had to register a developer account and authorize obscure permissions deep inside some Google permissions structure. I suppose if it's all straightforward yo
249.
by loudmax 2y ago
On some level, it's healthy to retain a sense of humility at the technological marvels around us. Everything about our daily lives is impressive. Just a few years ago, I would have been absolutely blown away by these demo videos. Six
250.
by loudmax 2y ago
The videos in this demo are pretty neat. If this had been announced just four months ago we'd all be very impressed by the capabilities. The problem is that these video clips are very unimpressive compared to the Sora demonstration wh
251.
by loudmax 2y ago
Mojo's developers claim that they intend for Mojo to be open source: https://www.modular.com/blog/the-next-big-step-in-mojo-open-... This is encouraging, and it seems that at least the core of the language is now
252.
by loudmax 2y ago
I think that depends what your expectations are, and what you mean by another AI winter. We're just scratching the surface of what's possible with the current state of the art. Even if there are no major advances or breakthroughs
253.
by loudmax 2y ago
If Tesla hadn't named their driver assistance system "Autopilot" they would have saved so much trouble (not to mention lives).
254.
by loudmax 2y ago
You're right, you can get an M3 Max MBP with 128GB, starting at $4700. My main point is that if your objective is dipping your toe into this, you can do it with smaller models for far less. That is a really sweet machine, but for the
255.
by loudmax 2y ago
I understand the preference to run models locally rather than in a rented space, but given the speed of development and the amount of money involved, you should have a clear reason why you need to run these models locally. As far as I kno
256.
by loudmax 2y ago
That is essentially what Peter Jackson did for the 2018 film They Shall Not Grow Old: https://www.imdb.com/title/tt7905466/ They used digital upsampling techniques and colorization to make World War One footage in
257.
by loudmax 2y ago
What I have in mind is a platform to run virtual machines at home. This is definitely a niche product, but there's a significant overlap between nerds running kubernetes at home and people who have influence over what server products t
258.
by loudmax 2y ago
Since Nirav Patel appeared on Bryan Cantrill's Oxide podcast ( https://oxide.computer/podcasts/oxide-and-friends/1632642 ), I've been hoping for a crossover product. I'd be interested in buying Oxide
259.
by loudmax 2y ago
These aren't North Korean entrepreneurs building businesses in a free market. Any enterprise in North Korea should be understood as slave labor to provide support for a criminal regime.
260.
by loudmax 2y ago
The Vernor Vinge novel Rainbows End famously presents a future in which people's interaction with the world is mediated by augmented reality via contact lenses. It's not presented as a necessarily bad thing, but who actually cont
261.
by loudmax 2y ago
The "in the wild" part of the title is misleading. These chips are being presented in a very controlled environment. An interesting aspect of Intel's design is they use Ethernet for connectivity. If they can get the perform
262.
by loudmax 2y ago
None of this reorganization matters as long as the executive suite remains the same. Google's problem is top-down. This rearranging the deck chairs isn't going to cut it. Much of Google's value proposition was in their abil
263.
by loudmax 2y ago
Disappointed to note that the 8k context length is far short of Mixtral 8x22B's 64k context length. Still, the published performance metrics are impressive. Kudos to Meta for putting these models out there.
264.
by loudmax 2y ago
You do want server-related infrastructure, at least so the clients can find each other. I think Matrix fits most of what you describe. I'd really like commercial Matrix offerings to become commonplace. I think there could be a market
265.
by loudmax 3y ago
To be fair, a lot of humans fail that. Including people that should know better.
266.
by loudmax 3y ago
I like that they say how the model was trained for 1.3 hours on 4 nodes of 8 x H100s. By my rough calculation, that should probably have cost around $100 or so. (At $2 per hour, x 8 gpus x 4 nodes). Not free, but pretty cheap in the sche
267.
by loudmax 3y ago
For those primarily interested in open weight models, that Mixtral 8x22B is really intriguing. The Mistral models have tended to outperform other models with similar parameter counts. Still 281GB is huge. That's at the higher end of wh
268.
by loudmax 3y ago
I haven't tried it yet, but if the model itself isn't impressive, that 128k context window is. That's the largest I think I've seen for any open weights model.
269.
by loudmax 3y ago
The 2024 Prius Prime has an MSRP starting around $33k. So about $8k more expensive than the hypothetical $25k Tesla, though it does have the advantage of actually existing today. It has about a 40 mile range on electric. When the battery
270.
by loudmax 3y ago
I wonder how this might affect GPU access to system RAM. Right now, people wanting to large language models on GPUs are constrained by the amount of VRAM the cards have. GPU access to system RAM is enough of a bottleneck that you might as

More ›