75 karma · joined March 21, 2024
There should be respect to the families that lost their property. But it's not insulting to suggest that NIMBY development exarcebates disasters like this. Heck if land use was as productive as it should in America most of those families wouldn't have a good chunk of their wealth evaporates as a result of a natural disaster, and rebuilding would've been hella more faster and cheaper.
The NPC live streaming is weird yeah but you cherry picked a trend and then make it about all TikTok. Literally hundreds of trends spawn up in TikTok every month and some of them are damn more authentic than whatever happens in IG reels. Some of the successful original trends even pick up in Instagram or YouTube.
I mean it reduces the communication overhead by more orders than DiLoCo.
Federated learning breaks the barrier to entry and expands the ecosystem allowing more participants to share compute and/or datasets for small players to train models.
DiLoCo introduced by Douillard minimizes communication overhead by averaging weight updates. What this article misses though is that despite this, each GPU in the distributed cluster still needs to have enough VRAM to load the entire copy of the model to complete the training process. That's where DisTrO comes in which even reduces further the inter-GPU communication using a decoupling technique (DeMo) that only shares the fast moving parts of the optimizer across the GPU cluster.
>And what if the costs could drop further still? The dream for developers pursuing truly decentralised ai is to drop the need for purpose-built training chips entirely. Measured in teraflops, a count of how many operations a chip can do in a second, one of Nvidia’s most capable chips is roughly as powerful as 300 or so top-end iPhones. But there are a lot more iPhones in the world than gpus. What if they (and other consumer computers) could all be put to work, churning through training runs while their owners sleep?"
This aligns with DisTrO techniques because, according to them it could also allow consumer devices like Desktop Gaming PCs to join the compute cluster and share workloads. Besides there's also an open-source implementation called exo that allows models to be split among idle local devices but it's only limited to inference.
Again might still be relevant since in the article it mentions that DiLoCo was able to make the model respond better when faced with instruction prompts or reasoning questions never encountered during pre-training. And Arthur seems to think test-time training will make his approach become the norm.
sources: DisTrO: https://github.com/NousResearch/DisTrO DeMo: https://arxiv.org/pdf/2411.19870 Exo: https://github.com/exo-explore/exo
It'll really take a special mind to think Community Notes wasn't a positive feature added to the social network sphere. Musk despite his schtick did very bold things that other platforms wouldn't think of doing, such as open-sourcing the recommendation system or recently suggesting the idea of optimising content with unregretted time spent that will reward healthy content and punish toxic content even if the two had the same number of impressions.
The overtone window is shifting towards a more open speech and less of self-gratifying echo chambers that promoted the toxic cancel culture.
Of course it'll not be easy so expect to put in effort and reps.
This video is just another degrowther canard.
Also I'd guess a lot of technical founders are more familiar with cloud/server-side than with dealing or delegating sysadmin taks that might require adding members to the team.
Once a model recognizes a weakness through reasoning with CoT when posed to a certain problem and gets the agency to adapt to solve that problem that's a precursor towards real AGI capability!