m.makes.musings

ml predictions

i've got too much pressure + self-confidence issues around posting things like this to my main blog, so it'll just go here.

the big thing i was bullish on around this time last year was multimodal and by... everything that i can see, multimodal is being slower to hit than i would've expected.

i think i underestimated how hard the problem was from both the research + inference side. but that's fine.

[that one company] only having 12 (!) people working on that area amazes me, but at the same time...

anyway.

i spent a little time looking at robotics companies just now which brought me back to this link that i'd seen earlier and idk exactly how right or wrong the following line is buuuut

According to media reports, OpenAI is now on the verge of developing a host of multimodal large language models for robotics use cases

that feels right to me.

Specifically, it's related to the post I made a few days ago. There's a severely missing notion of time with current LLM architectures and tasks and datasets and...

Robotics sort of forces tackling all of this.

it's funny - while i was interviewing around the chats of the... not-as-on-the-forefront companies were mostly asking about handling multimodal web data.

oai's got to have already done all of that like, a year plus ago.

wonder if this is gonna highlight that much more the style of death that some of the other llm cos are gonna end up having.