So cool, inspiration for many.
K-Scale Labs

So, just a thought, I think the selection criteria should at least balance representatives from the corporate world with those from academia.

while XAI does not evolve accordingly

Reconfigurable Dataflow Architecture (RDA) is going to rival LPU Inference Engine?

This guy ain't playing around.

The new phi-3 with 3.8b running locally on an Iphone 15 generating 12 Tokens per second.

I was not expecting for this one. Phi-3 with 3.8b beating Llama 3 8b instruct on most benchmark metrics. Reading their technical report right now: export.arxiv.org/abs/2404.14219

This is very cool, even if in practice it ends up having a lot of flaws

Damm, how can this even be allowed.

Concerning

What a pleasure it is to watch

Planning with a good model of the world and of otter members of the group.

Excellent website to compare API providers for LLMs,
including price, throughput, variance and latency.
Here for Llama 3 8b:

artificialanalysis.ai/models/llama-3…

It's really impressive this real-time feature.

However, I would much prefer the kid to imagine without it.

I haven't fully processed it yet, but it somehow seems to alleviate the integral role of the imagination and passively relying on the visual system.

In Human Preferences Evaluation:
Microsoft's WizardLM 2 70B shows signs that it could be better than Llama 3 70B. Check the comparison against Mistral for both.

Currently the page is out - they are running a new 'toxicity testing'
web.archive.org/web/2024041601…

Andrew Ng They held it on purpose 😄
Happy birthday, Andrew