I made an original game with Claude AI + Gemini. And it's awesome!
It is called: Yarn snake.
https://t.co/B5fV9bRfHJ
#claude#yarnsnake#originalgame#gemini
@v1ctus17@OfficialLoganK@mercor_ai Governance, itsm, ITIL processes, lots and lots of files.
If I open up a meeting at Google because I am at a war room.
I activate real time translations, recording, transcriptions... Gemini will do the rest.
Drive, notebooklm, slides, video instrutions, roadmaps, everything
@v1ctus17@OfficialLoganK@mercor_ai You cleared haven't tried Gemini 3.5 flash long enough.
The way workspace acts and reacts to itself integrating google meetings, notebooklm, docs, sheets with just one single prompt after the i/o event is mindblowing.
I handle file overload, company size.
AGI feelings here...
@OfficialLoganK Final: Everything else, graphs, critics, RAG is scaffolding around that one broken signal.
One example is a solution layer for this...
RAG + Knowledge Graph anchoring
Overconfidence,
RLCR / Calibration-aware training
Unchallenged errors
Adversarial critic model... Etc...
@OfficialLoganK Part 2: ...the world, and start giving it the senses to experience it.
I don't think a new world model wikl fix this. Before AI gets its "senses" it will need better rules, confident wrong answer costs more than "I don't know."
Retrain the incentive.
Nothing changed latelly. Perhaps Mythos and chatgpt Cyber could change my mind, but I haven't had the access to those yet.
So, we're still at 87.5% completion towards AGI.
Next update, as said before will only come out after the first OS and/or AI device comes to public.
#agi
Like I promised, I said that I would only place a score based in the AI progress for reaching AGI, only when the speedrun towards AGI would calm a bit.
We are at 75% of progress towards AGI
#agi#machinelearning#deeplearning
Introducing SubQ - a major breakthrough in LLM intelligence.
It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),
And the first frontier model with a 12 million token context window which is:
- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus
Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).
Only a small fraction actually matter.
@subquadratic finds and focuses only on the ones that do.
That's nearly 1,000x less compute and a new way for LLMs to scale.
Nobody, and I repeat NOBODY, will be able to control the next great phase of the AI age. Not even the CEO's, not billionaires, nobody. History is following exactly all of scifi prophecies. And even if they succeed, it won't last long!
Dario, Altman, Google, Elon, Ilya.
No one!
@LukeberryPi A resposta pra isso é mais óbvia do que parece.
Pra quê vou usar ultra pro max deep thinking mithos plus mode pra traduzir de árabe pra português "a maçã é redonda"?
Isso é um exemplo irônico, mas você tem melhor aproveitamento dos llms se poupa limites de uso.
@VraserX We should stop listening and paying attention to what billionaires say. Half of what they say is to enrich their own enormous and senseless wealth, and the other half is to strengthen policies that strengthen their empires.