nVidia’s new Digits workstation, while expensive from a consumer standpoint, should be a great tool for local inferencing research. $3000 for 128GB isn’t a crazy amount for a university or other researcher to spend, especially when you look at the price of the 5090.
In fairness, unless you have about 800GB of VRAM/HBM you’re not running true Deepseek yet. The smaller models are Llama or Qwen distilled from Deepseek R1.
I’m really hoping Deepseek releases smaller models that I can fit on a 16GB GPU and try at home.
Linux is a lot, lot, lot easier to use now than the 90s.
Mobile phones in the era before smartphones had cameras, email clients, games, music players, and even web browsers. They just weren’t very good at those functions and their core feature was being a phone for voice calls. Texting was barely a feature on some of them (the first camera phone in the United States, the Sanyo SCP-5300, didn’t have a two way text messaging client - the user had to go to a website on the phone to send texts, which was inconvenient even on a 1xRTT 3G connection.)
The e-ink phone seems closer to a dumbphone than a smartphone, IMO, largely because it lacks access to an app store.
Source: I sold mobile phones before smartphones and during the early smartphone years (BlackBerry and Palm Treo, for example.)
Edit: calling it a feature phone instead of a dumb phone might be more accurate.
This also grinds my gears, especially after so many officers die during the course of the series. So many junior officer ranks are time-defined promotions (in real life) and even if they weren’t time based, slots opened up due to all the people dying.
I hate to break it to you, but if you’re running an LLM based on (for example) Llama the training data (corpus) that went into it was still large parts of the Internet.
The fact that you’re running the prompts locally doesn’t change the fact that it was still trained on data that could be considered protected under copyright law.
It’s going to be interesting to see how the law shakes out on this one, because an artist going to an art museum and doing studies of those works (and let’s say it’s a contemporary art museum where the works wouldn’t be in the public domain) for educational purposes is likely fair use - and possibly encouraged to help artists develop their talents. Musicians practicing (or even performing) other artists’ songs is expected during their development. Consider some high school band practicing in a garage, playing some song to improve their skills.
I know the big difference is that it’s people training vs a machine/LLM training, but that seems to come down to not so much a copyright issue (which it is in an immediate sense) as a “should an algorithm be entitled to the same protections as a person? If not, what if real AI (not just an LLM) is developed? Should those entities be entitled to personhood?”
I paid for 1Blocker (https://apps.apple.com/us/app/1blocker-ad-blocker/id1365531024) since it runs locally on my phone to block ads as well as trackers from other installed Apps, whether on my home WiFi, other WiFi, or cellular data. There’s a whitelist feature and the filtering can be tailored. I’ve been using it for years (before I got my current phone) so I can’t comment on battery life but I’ve never noticed it to be bad.
It does break some apps if they’re not whitelisted but that just gives me more reason to decide if I really need the app. It’s also possible to temporarily disable the blocking (5 minutes, 30 minutes, etc) which helps with logging in (since so many login services involve trackers).
I know in my state NA beer is still considered to be beer. Have to be 21 to purchase, need a liquor license to sell it, etc.
As I recall it’s because the definition of beer in the state statute defines beer as a fermented drink with certain criteria like wheat, barley, hops, etc. as ingredients, and alcohol content not being one of the criteria.
It actually meant a local sober bar (a bar-like place that only serves NA drinks) had to go through the hassle and expense of getting a liquor license.
If Apple chose to use a 16-bit integer for the index, they would be limited to approximately 65,000 photos. 32-bit would be about 4 billion. Photos likely uses a 32-bit or even 64-bit integer index value to make it a non-issue.
There could be issues with having too many thumbnails loaded in memory. Maybe some other value overflows when scrolling through (or just loading) an enormous library. If Photos also creates an index for searching that could also create headaches. If Photos were a simple file explorer having a million files shouldn’t be an issue but it’s more than that, so it’s good to know it seems to be performant with even 1 million photos.
Lucky you! I need to check my university’s current GPU power but sadly my thesis won’t be needing that kind of horsepower, so I won’t be able to give it a try unless I pay AWS or someone else for it on my own dime.