
> Crouching tiger, hidden layer(s)
Argh!
Next you'll be telling us that the CEO/Founder's name is "Hugh"
The last thing the world needs is yet another disinformation machine.
Barely a week after DeepSeek's R1 LLM turned Silicon Valley on its head, the Chinese outfit is back with a new release it claims is ready to challenge OpenAI's DALL-E 3. Released on Hugging Face on Monday amid an ongoing cyberattack, Janus Pro 1B and 7B are a family of multimodal large language models (LLMs) designed to handle …
Amy Castor and David Gerard have a few interesting things to say about DeepSeek's claims.
Most interesting to me were the suggestions that DeepSeek is using just as much computing power as anyone else, they just can't admit to it because that would imply they've breached export controls on Nvidia GPUs, and the suggestion that OpenAI is pushing the DeepSeek hype as a scare tactic to attract more funding (because they're rapidly running out).
Perhaps a little bit conspiratorial, sure, but at least they list their sources.
The first one can be easily tested and will be. The second one would crash and has crashed Nvidia stock by 600bn. I know you can short stock but I really can't see someone doing that. Clearly those in the business of AI knew or had an idea this was coming as it's not like it's just magically appeared. What this does do though is put into question some of the funding floating about because it's not longer a relatively safe investment. It will be interesting to see what the orange mans puppeteers get him to do though I don't think there is much they can do. If America wants to stay ahead of the game it's got a lot of work to do now.
The game is creating something better. If the same thing can be done for less and with less then the game now is to actually create something better that uses more and costs more. That was the model all along wasn't it?
The logical target now is the babel fish. Which will finally prove that man is his own god and end all religious wars on the planet delivering world peace.
>Perhaps a little bit conspiratorial, sure, but at least they list their sources.
(here's an actual conspiracy)
The smart money has had months to reposition holdings before this week's inevitable adjustments.
Got to admire the front of Altman etc over the past few weeks (and months) since they certainly knew what 'China' was up to by last summer at latest.
https://web.archive.org/web/20240816071628/https://36kr.com/p/2872793466982535
Symbolising the transition from one year to the next, hence January, looking backwards to the time just gone and forward to the future. In this case backwards to a time when the money was flowing into certain companies like a river, and forwards to a time when all that money just evaporated into thin air. This is looking more and more like possibly the largest, most meticulously planned trolling operation of all time.
'This is looking more and more like possibly the largest, most meticulously planned trolling operation of all time.'
Maybe trolling but I wouldn't say well planned. I have energy company shares which grew and then dropped 30% above and then back to the intrinsic value of the company over the last month.
That's a bubble.
Far more damage would have been done if that bubble was allowed to grow to the point major investment was actually made (dot com bubble anyone?).
That's poor planning for a trolling operation so adds weight to the authenticity of the claims. The perpetrators of the troll would gain more if they had waited.
One problem is that the markets have created a "there can be only one" attitued about tech. Their model reaquires endless exponential growth which, ultimately, excludes dividing the spoils between companies. Competition used to be seen as a good thing to spur innovation.
Do tell, how does one "steal" something and the something that was "stolen" is still there afterwards?
Could it be, maybe, because no theft (as in the legal definition of the word) takes place, but rather: "someone is looking at something someone else put online, but with a program rather than eyes" is what is really going on?
Because if so, well, then there doesn't really seem to be any theft. At worst it is something to do with copyright, and now we face several interesting questions, such as "is a ML model a derivative work of art" and "what countries laws do apply".
Simple question: If someone takes a ruler and set square to a museum, and takes statistical measurements of the images, and then develops a statistcal model on how the images work, and publishes that model...where exactly does theft, or copyright infringement take place?
Quite. These things learn (AIUI - could be completely wrong) in a similar way to humans - lots of repetition and lots of source data. Just faster. So the results are similar to what you might be from an average human trained on the same data. Do we accuse humans of theft when they draw a picture that looks a bit like someone else's picture? No, but copyright maybe infringed.
Same with text, there are only (according to Booker) 7 plot lines - so all books have their basis in those 7 plotlines, or are derivatives of those plotlines. Writers take those plotlines and build around them in their own style. Is it any different with AI?
Dunno, but it's all quite interesting.
My hope is that all of the money for investment disappears overnight, as people realise these things can be spun up relatively cheaply, and that with so many potential competitors, the expected profits will be minimal (or more likely negative!).
With that investment money gone, the models will stay dumb, regular people will realise there dumb, and stop using them. Other people will determine ways to tell the difference between LLM output and a real persons output, and so make it obvious whose using LLM's.
LLM's will find there niches in places liking helping coders, or speeding up analysis, but will disappear from the creative spheres, as well as science (outside of some analytical assistance).
The copyright cases against the big names will lead them to paying out millions for theft, and destroying large amounts of their data. In addition, this will hopefully have the effect of making their techbro bosses unemployable, or at least unable to rake in investor cash for their next crappy attempt to fleece the public.
And with all of this crashing back to the ground, the latest bubble is burst before it can go forward with destroying the environment (through energy emissions) and our creative landscape (through destroying the creative industries with their theft)...
I guess a man can dream...