Comment: Initial grayskull impression
Transcript: First impressions of the Ten's Torrent Gray Skull. It's a lot of work and it runs Whisper very slowly compared to my GPU or M1, so... It's early hardware, what can you say? That's kind of what I would expect from early hardware, but also I was really hoping for just like, you know, smashing performance out of the box, so I could just kind of run Whisper all the time on this card. But it doesn't seem like that's going to be the case, so yeah, we'll kind of just have to keep a tab on it and see how it continues to evolve, go from there.
80.53% similar
The speaker is focused on their API project and mentioned the stable diffusion model. They have also worked on running and testing various local language models, including Whisper and Orca 7 billion. They are curious to wire the models as a pipeline step and compare the output with GPT. The speaker is unsure of the success of their API project and the effectiveness of the language models, but they express eagerness to explore and experiment further.
80.38% similar
The individual is contemplating the potential local costs associated with using technology, such as Whisper, which may not be fully captured by the initial price. The economies of scale are considered, highlighting the complexity of determining the overall impact. There is uncertainty regarding the ultimate deployment of such technology, particularly in relation to large language models. The individual expresses a desire to further engage with the technological community and contribute to solving challenges faced as a developer.
The speaker reflects on the customer-focused approach of TSMC and its importance in business, while also discussing a personal struggle with balancing self-focus and advancing the interests of others. They find satisfaction in moving the ball forward for other people and are working to create tools that fulfill initial promises and improve ease of use. The "burrito project" mentioned seems to involve creating fundamental tools, exploring AI, and making computer tools more accessible, highlighting the magic and potential of fetching personal data with simple requests.
I had an excellent call with Jordan, which I'll discuss in more detail later. Unexpectedly, I succeeded in quantizing lava, and the initial results look promising despite the script being rough. It seems feasible that I might run something akin to GPT-4 style vision captioning locally. I'm excited to upload these quantized models to see the outcomes and I plan to head to Jordan's house to upload more models soon.
I intended to note last night that I might be overly focused on the technology aspect. Jordan, who has confirmed my significant progress, joined me for burritos and I updated him on my work. I value Jordan's assistance and am grateful for his help. My aim is to get the product into users' hands for feedback, despite any associated anxiety.