PhD candidate @MIT | AI × robotics × aerospace | building autonomous systems you can trust your life with | 🇺🇸
Joined September 2016
- Tweets3.4K
- Following3.2K
- Followers1.2K
- Likes55.4K
Blake Edwards retweeted
Guys, I beg you to start being the person you wish to be. This study found that just 5 minutes a day of imagining your best possible self, for two weeks, made people noticeably more optimistic about their future. Picture who you want to be, then go act like them today.
The biggest risk right now is that the West fails to rise to the challenge and win the AI race.
Our AI leaders’ repeated PR disasters threaten to provoke a political backlash that could completely undermine the progress we need to succeed.
Blake Edwards retweeted
Where computational irreducibility makes particle kinetics infeasible, it gifts us fluid mechanics to compensate.
Where it makes neurophysiology infeasible, it gifts us psychology to compensate.
Mechanistic interpretability is infeasible. What have we been gifted to compensate?
Some of y’all have not studied the implications of computational irreducibility, and it shows…
Local determinism wrt nontrivially interacting rule sets does not help predict the “long term” behavior of a computationally irreducible system, and when it comes to AI the iterative frequency is such that “long term” is not actually that long from the human reference frame.
This is as true of our current neural nets as it is of the initial cellular automata in which computational irreducibility was first discovered and studied.
It’s also what Anthropic was (re)discovering wrt their escape cone diagrams wrt agentic unpredictability over long-horizon work—especially work with somewhat ambiguous constraints and success metrics.
Except that they got the cone geometry wrong, as the region of an agent’s potential state space trajectory looks more like an exponential horn.
“The pkill -f pattern matched my own shell command and killed it before the probe ran.”
Every. Single. Time. 😂 Must be really etched into the training data.
As models become increasingly powerful, generalizable, and economically valuable, we will likely see the emergence of highly optimized model-specific integrated circuits (MSICs), as the economic benefits of designing specialized hardware for individual models begin to outweigh the costs.
The resulting gains in computational efficiency and inference speed could be so extraordinary that they fundamentally defy even a moderately technical observer’s intuition about what is technologically possible.
Blake Edwards retweeted
We are now at "the fuck" levels.
"Our initial evaluation suggested that GPT-6-Astra's score was in the late eighties followed closely by Opus 5.5. But, as we started to investigate the mistakes that the models might have made, the models, especially Astra, insisted that they were right and that it was the expert-supplied answers that were erroneous. When we went back to the experts it turned out, in multiple instances, that that the experts found some subtle errors buried deep inside their calculations. As they revised their answers, Astra's score started to inch up and it finally ended up with 98% (49/50)."
Blake Edwards retweeted
String theory is what happens when an intelligent system is starved of sensory input and starts hallucinating. Intelligence needs grounding.
For this reason, intelligence is practically never the bottleneck to solving impactful problems. The bottleneck is iterated contact with reality.
Blake Edwards retweeted
Ruthless with elimination.
Violent with execution.
Resilient with setbacks.
Relentless with iteration.
Unyielding with standards.
