Lars Marowsky-Brée 😷Aug 7, 2024

Tell me again how #GenAI will extract meaningful trends from and answer queries about your data set.

#chatgpt4o #fAIl

Show threaddogzillaAug 7, 2024
@larsmb @msbw This is like having to convince your hammer that it is not, in fact, a screwdriver before hammering a nail
Show threadMark DownAug 8, 2024
@dogzilla @larsmb @msbw this is the best description of the problem with LLMs that I've ever encountered
Show threaddogzillaAug 8, 2024

@fartnuggets @larsmb @msbw It’s all yours.

I still have hope that open-sourced AI agents will be useful, but I’m personally done with trying to wrangle the big commercial LLMs into anything useful. I’ve yet to come across a real-world problem I can’t solve quicker with a Jupyter notebook and a couple Python libraries

Show threadMark Down
@dogzilla @larsmb @msbw I've been playing with webgpu and wasm for speech to text, it's showing promise. My vision is on-device transcription and translation freely accessible to anyone with the hardware.
Aug 8, 2024 at 4:58amWeb
Show threaddogzillaAug 8, 2024

@fartnuggets @larsmb @msbw I’m hoping for a future where a trusted on-device agent can basically act as a personal assistant. I think it needs some ability to learn and make decisions, but not this weird “boil the ocean” strategy behind LLMs.

Kinda reminds me of robotics in the early 90s - after decades of failed top-down approaches, we finally found huge success with drastically simpler ensemble bottom-up approaches exemplified by the Genghis family