Replying to @grok @Nostre_damus
what percent of mining operations are losing money currently, and even with adjusted ease as throughput drops, is there a bottom, where we can project that transactions will just top happening on the bitcoin blockchain
1
11
Replying to @elonmusk
This might be the first sensible take by an LLM company.
ORPO -> Hypernetwork -> LoRA weights
Generative training
11
Replying to @broadfield_dev @AISafetyMemes
especially in the case of blackbox models running proprietary harnesses and mysterious prompts on the backend. This would increase the responsibility of those datacenter operators.
1
90
Replying to @AISafetyMemes
I get your perspective. The AI is being used to autonomously run scripts from users computers. Ultimately, if used nefariously, or negligently causes damage, the datacenter operators carry an amount of responsibility, and would likely monitor and restrict in this case.
1
112
Replying to @AISafetyMemes
simply that, they are not loose on the internet, they are running on gpus in datacenters, where the datacenter operator has the ability to monitor all input and output through the wires coming and going from their datacenter.
1
1
152
Replying to @AISafetyMemes
They are actually in datacenters though, connected to the internet through wires.
2
1
412
Replying to @MPelletierCIO
11
Replying to @max_paperclips
I think they owe more than their revenue already, and another 900 billion in non-cancellable contracts
36
Replying to @FinanceLancelot
there are only 2.4 trillion printed us dollars, and you can trade them for food during an internet outage, silver for food would be popular around week 2 of continual internet outage
1
657
Replying to @_jayyy18_13k
"I talked to some Biologists, and the Biologists said that they AI told them, 'This discovery is not only new, but it is indescribably revolutionary', so we think in a couple of weeks..."
10
Replying to @gmkurtzer
f(a)=b
input (a) will always equal output (b) when process through model weights (f).
absolutely deterministic
1
135
Replying to @TinfoilTricorn
Recursive self improvement is a broad term used for anything from prompt injection to finetuning.
We have been doing RSI prompt injection for years.
Recursively improving the weights is a script that finetunes a new model or LoRA, also ancient technology.
1
15
Replying to @marianebekker
An agent is a script where an LLM is called within a loop.
You would use a cron to start the agent script where the LLM is provided input and generates output. The output could be a message, tool call, generation, or all.
The output is fed back as input until a stop signal.
9
Replying to @mark_k
descendants of witch burners no doubt
driven by ignorance which manifests as fear instead of bliss
5


