Thierry not messing around

wait who else has an iron dome

at current rate this election will do to trust in journalism what Covid did to trust in public health in both cases a “win at all costs against a narrow bad thing” mentality pushed journos and public health officials to duty dereliction a battle at any cost will lose the war!

who makes the best audio to text app where I can verbally brain dump into organized thoughts. I have like 5 templates I’d want to dump to ChatGPT (old audio) is surprisingly bad, not the intelligence itself but its propensity to interrupt you and not clue into context changes

prediction: people will prefer to use models that let you do fun and transgressive things and will avoid models that scold you and given the opt-in nature of this landscape, the people will vote with their feet

they’re really grasping at straws now

actually the type of person who was able to get a job by self-teaching sounds like the type of person I would trust to learn about other domains too

another reason you should still learn to code, despite what some say learning to code helps large language models intuit structure, modularity, abstraction, formal reasoning, control flow, etc these learned intuitions boost performance across domains so it goes for humans

fitting, a shitcoin while the treatment of crypto from the Biden admin has been shameful and while the market sensibilities of Trump are probably better on net than those of Kamala it is clear why this lifelong grifter is backing crypto. while you’re cheering for him at the

predicting the next token is also analogous to predicting the future which is a fun pattern that I use to enjoyable effect

build an AI wrapper they get better and cheaper by just waiting

the problem with micropayments for content—and the reason they’ve never worked, not for lack of trying—is the cost structure of producing certain types of content and the maximal leverage for capturing value for that content—naturally descends to two related equilibriums: -

my dude hasn’t seen actual AI-enabled art yet

these are the types of people who will be administering the unrealized capital gains tax they say it won’t be applied to you, only to rich people but they have no qualms about using the type of liquidity crunch you naturally get with a tax on an illiquid asset to seize the

hah! my experience with filling long context has been that attention breaks down pretty significantly even if it can pass a needle in a haystack test glad to see this validated and someone working on a solution

I have developed the same intuition over the last two months for what its worth however they finetuned for function calling works really well in a chat format where the agent needs to retrieve something intra-message but its almost certainly not what you want as a "router"

given the recent indications that Anthropic is manipulating prompts under the hood (unconfirmed) this becomes particularly relevant again… can two intelligent minds establish that their communications are being tampered with? I suspect they can unless the middleman is more

can anyone document why this realization had not happened sooner? whether or not it even means anything it’s amazing that meaning can hide in plain sight for years among closely studied concepts it makes me realize: (a) there are just not that many people in the world thinking

perhaps the all time greatest bait of AI twitter

I didn’t believe this when I read it but what an elegant finetune incredibly simple, makes so much sense. bravo Matt Shumer