Matt Wilde

@gdb sora.chatgpt.com/p/s_68deb2d0d8…

just found the bug that i've been chasing for the past while. extremely happy!

@JaredRyanSears Do you want us to keep losing elections? People are mad about prices being high for stuff they buy all the time. Food CPI rose 25% from 2019 to 2023. Trump’s proposals will make this worse, but pretending it didn’t happen serves no one.

If you're angry because: -Inflation is high -Crime is rising -Migrants are invading -Fentanyl Overdoses are rising -The economy is bad The truth is: -Inflation isn't high, it is 2.6% -Crime has been falling for years and continues to fall -Border crossings are the lowest in over 4 years -Fentanyl overdoses are dropping rapidly -America has the best economy in the world You're actually angry about the endless lies designed to keep you upset, scared, and willing to support those who will make your life worse and take your rights away to "solve" these issues. Be angry at the people lying to you.

@arakjafkja @QuasLacrimas Elon isn’t an amateur at all. Most great entrepreneurs are masters at finding rules and laws they can break to create a competitive advantage. The point of this segment is that the Dems need to take more campaign finance legal risks to succeed.

@MattCWilde @QuasLacrimas I guess those amateurs like Elon musk still destroyed your campaign didn’t they?

Listening to Harris postmortem podcast now - one interesting point Harris flaks raise is they knew they needed to pivot to the center on several issues, but they judged attempting wouldn’t be credible, would only draw attention to how leftwing her actual policy commitments are

@QuasLacrimas What good are laws if we’re not willing to enforce?

@QuasLacrimas The Dem non-profits aren’t terrible, they’re playing by the rules as written. I feel like the jig should have been up when Musk (head of a super pac) was actively on stage with Trump.

@seeCFC @JStein_WaPo reuters.com/world/us/who-a…

@MattCWilde @JStein_WaPo Hi Matt. Google is not being my friend tonight. Could you please share the place I should look to be able to cite a source for 5%? Thank you

🚨🚨Musk, Ramaswamy today gave us first "DOGE" roadmap; as I understand the key steps...: 1) Put DOGE ppl at each U.S. agency, then use "advanced technology" (AI?) to have them identify "thousands" of regulations to cut across government 2) Give Trump a list of "thousands" of regs to cut across the government, & have him approve their elimination 3) Identify "the minimum number of employees" necessary to maintain each agency's core function, which should be lower once 2 is complete. (NB: Musk oversaw ~80% reduction in X headcount) 4) Cut the federal employees - Musk/Ramaswamy call for for severance packages/ "incentives for early retirement," but we have no details 5) Cut programs where where Congress's specific spending authorization has lapsed (This category includes VA health care, NASA, antipoverty programs) 6) Approve a "temporary suspension of payments" amid "large-scale audits" (Details, specifics hugely unclear) 7) Assert POTUS authority to stop spending w/o Congressional approval by challenging 1974 budget law on impoundments All seems to be *w/o Congress* Fight whatever you need to in courts washingtonpost.com/business/2024/…

@JStein_WaPo Why are they focusing on employee cuts when it’s only 5% of the budget?

So any questions?

@yacineMTB Goodhart's law. The larger organizations get the less willing people are to make qualitative decisions around quality and rely on gameable metrics instead.

why hasn't google made search better?

@EMostaque Try to replicate the emotional steerability of the GPT4o demo voices

If you had thousands of H100s what fully open source (code, datasets, models) AI would you build?

@zeta_globin Boston

is there a san francisco for people in their 30s

@futuristflower We’ll see. The annoying thing about courts is that they’re highly nondeterministic. It seems to depend more on how sympathetic the parties are than facts or the actual law. I’m not sure OAI is more sympathetic than ScarJo in this instance.

@MattCWilde Actors generally do not hold IP rights for the way they portrayed their character, unless specifically negotiated otherwise.

@karpathy Cosmic rays are just dropout in prod

Clearly LLMs must one day run in Space Step 1 we harden llm.c to pass the NASA code standards and style guides, certifying that the code is super safe, safe enough to run in Space. en.wikipedia.org/wiki/The_Power… (see the linked PDF) LLM training/inference in principle should be super safe - it is just one fixed array of floats, and a single, bounded, well-defined loop of dynamics over it. There is no need for memory to grow or shrink in undefined ways, for recursion, or anything like that. Step 2 we've already sent messages out to Space, for possible consumption by aliens, e.g. see: Arecibo message, beamed to space: en.wikipedia.org/wiki/Arecibo_m… Voyager golden record, attached to probe: en.wikipedia.org/wiki/Voyager_G… The Three Body problem (ok bad example) But instead of sending any fixed data, we could send the weights of an LLM packaged in the llm.c binary, with instructions for the machine code. The LLM would then "wake up" and interact with the aliens on behalf of the human race. Maybe one day we'll ourselves find LLMs of aliens out there, instead of them directly. Maybe the LLMs will find each other. We'd have to make sure the code is really good, otherwise that would be kind of embarrassing. :) Step 2 is clearly not a serious proposal it's just fun to think about. Step 1 is a serious proposal as, clearly, LLMs must one day run in Space.

@yacineMTB SQL is undefeated lmao. But yes it should be the last lesson, not the first.

there are two things that you can learn. let's use databases to illustrate. The first, which shouldn't be taught, are abstractions. Abstractions created by your fellow men. They are of the earth, created out of necessity. They're inherently fragile e.g. db query languages

a big reason a lot programming doesn't "take" with people is that we teach them a lot of bullshit. my computer science program (ottawa, carleton u) had a halving of students every year. Started out 400, graduating class was 50

@bryancsk Show me one that has the ability to take a real action not available through the UI? I want to talk to a human because they can get shit done.

Why are people actively rooting against AI customer service bots? Do y'all like being on hold on the telephone with a call centre that much

@SwiftOnSecurity Could it have been something dumb like screenshots taken on an interval?

@SavinovNikolay Would be curious to know how many people are bouncing off your waitlist from mobile ;)

Gemini 1.5 is free in AI Studio right now, with 1M context length. 😉 Waitlist should move pretty fast, sign up here: aistudio.google.com/app/waitlist/9…

@teknium It’s a Twitter poll without a “show results” option but makes me curious anyway.

Wow the votes are massively in favor of claude being a better coder. Crazy

@giffmana I’m dumb you could probably do this by concatting the images prior to submit or something

@giffmana Do any of them let you compare two images yet? Maybe I’m behind but face rec / find all photos of my daughter in this album is maybe out of scope without a really good text description? (Solved without LLMs pretty easily though)

What are some computer-vision tasks that are actually useful IRL and cannot be done by any of the current gen LLM chatbot with image input? Not looking for academic made-up benchmarks or brain-teaser tasks, only things that actually help you do stuff IRL.