The deeper I get, the more aware I become of the wide gap that exists between the hype and the state of reality.
5
Some months back I gave a pitch at wema
Didn't make the cut
However work continued behind the scenes on the tech
My understanding of where this could take us has deepened however at this point we're still largely a loss leader
Yet for science we must prevail
Something I find myself interested in is emergent behaviours
When you simply let something be, within a controlled environment you can sort of predict its reaction to scenarios, throw in variations and it might learn some new methods, ones you could not have taught it
5
To me anything is a problem that needs to be solved, to the point that my personality and behaviour are all things that need a solution.
This trait has led me down some really good paths, I won't lie
Though it's not very energy efficient
Something I find myself interested in is emergent behaviours
When you simply let something be, within a controlled environment you can sort of predict its reaction to scenarios, throw in variations and it might learn some new methods, ones you could not have taught it
3
I'm not the mathiest in the room but my model is and I'm able to reason about systems
Most outcomes follow a determined process
Solve the math behind that
Have a smaller model do the math
Theoretically you've just distilled the model (loose ends here but who cares)
1
11
Voice model WER was too high
ASR has a better chance though I'm working to see if we can get same results on much less params
5
98.7% cache hit actively
Overral 94%
Total tokens 3 billion +
Within 4 weeks
Pretty efficient if I had to say so
1
11
Nothing still comes close to Midjourney
It produces actual art
The reason @SpaceXSI laid off the Grok Imagine team is because coding models will soon overtake video models for video generation.
You saw glimmers of this with some of Fable 5.2’s outputs. It can generate videos by writing code that are better in some cases than native video models.
I’m starting to think every task of the future will be solved by code.
21
Something I find myself interested in is emergent behaviours
When you simply let something be, within a controlled environment you can sort of predict its reaction to scenarios, throw in variations and it might learn some new methods, ones you could not have taught it
There's a very rough way I've found to go about solving models
It doesn't involve a lot of reading papers, the llms can do that much quicker and better, doesn't involve a lot of math proofs, again llms.
It's just doing the thing, hey Gemini train a model for me.
It works, try it.
1
14
Just remembered when openai was sending awards to token maxers
10bn 20bn token spends
Seems like such a long time ago
1
31
The bar is so high these days
Hey guys! I'm looking for a remote, full-time Software Engineer role in backend, frontend, or AI.
I've built real-world products processing millions of naira and serving hundreds of users.
Portfolio: samueltuoyo.com
Retweets appreciated. 🙏
1
24






