coderatlarge

how accurately can a model do lip reading these days?

well it’s widely understood that any large scale social media account is managed by a team and not the individual being represented and we seem broadly ok with that. what those teams produce is often vapid manipulative corporate bs which i don’t find any less offensive for it being typed by humans earning a salary as compared to tensors assembling copy via gpu. the current anti AI sentiment strikes me as a bit of a high-minded pose when the actual underlying behaviors have been in place for a while.

i personally prompt to produce a baseline then re-prompt to tweak then re-re-prompt to re-order, then re-re-re-prompt to expand certain parts, then re-re-re-reprompt to properly cross reference internally , then re-re-re-re-reprompt to proof read and add missing references . so there is no single prompt that produced the document and if i were to start from scratch the path would likely be different.

i wonder if you’re equally offended by a public figure hiring young people to ghost write for their social media accounts or their books etc

prompt: if a family has four boys, would a bayesian conclude that their next child would most likely be a boy?

fable taught me about the beta binomial and large observational studies that came down on both sides of the question about whether per family births are truly binomial. it also told me about countries like the uk and uae that are inching towards national genetic registries that might answer such questions definitively in time. as well as the efforts in Cyprus in this 80s to reduce beta thalassemia through voluntary testing of couples pre marriage.

curious to hear about your findings!

i’m personally deriving a huge amount of value from the custom materials fable is assembling for me. for example i asked it to write a focused expository math paper on reed solomon to accompany an implementation module that it wrote for me. it’s remarkably useful to steer it to create graphs and diagrams of exactly how you like the material presented. or the bibliography researched and cross-linked with the body or the order you want your questions addressed.

it also researched vision correcting displays for me and i can finally put that idea to bed - i was never really going to pick up an optometry textbook tbh. plus it was able to pull together a bunch of geometric and physical context about light and the eye plugging exactly my personal knowledge gaps.

in general i suspect these materials might not be that interesting to others because they are so custom to my learning style and personal needs and preferences.

these are usually not one shot documents but rather many prompts deep before i get something I’m willing to sit down and read or study. but dramatically quicker than assembling it myself from primary sources. i wouldn’t say it matches master expositors but then they’re not available to write on any topic i happen to need right now.

plus I’ll just have a live voice discussion with the system when i go for a walk and there are still things bothering me on a topic. it takes a little patience but if i’m in the mood it’s amazing.

i generally find that it can help track down specific references if i suspect hallucinations. but especially on factual topics my experience so far has been extremely encouraging.

except for most other companies in the world.

breaking some of these systems that humanity has been banging on for decades would be an elegant proof that the llms have outsampled us decisively. one word at a time, which is how we write too.

i meant people who enjoy the idea of software even if they don’t necessarily need or enjoy the resulting software itself. it’s possible that some of the spreadsheet hacking has a similar flavor but lacks the natural language hook.

i would make a similar argument for entrepreneurship and most things: do it only if you can’t bear the thought and reality of doing something else.

i consider it as a very specialized subset (subgenre) of video games for a very narrow audience

i use it primarily to enforce my personal conventions that are not super commonly followed in the wild. structure of comments at the top of .h , structure of bibliography in .tex, etc etc things that i would otherwise have to repeat periodically.

can i provision from within china or do i have to do it before i go there?

looking forward to it!

how do you plan to publish?

do you offer non gfw esim for china?

claude helped me build a functional visual modem (using the cameras and microphones/speakers) for mac/iPhone in about a day and <100 prompts. i’ve worked on mobile apps once 10 years ago and have never built one from scratch. i've built a handful of UIs in 20 years on different platforms. my actual work was mostly in systems software. so it would have probably taken me weeks to months to build this from scratch. it would have cost substantial money to outsource this and iterate on ui and core algo as i did.

i've been curious about the feasibility of this concept for quite some time and now i have an actual example to study and measure and hack on, for a fraction of the cost of my pro sub. it wasn’t trivial, i doubt a non-swe could get it done in that time, but i'm a pretty happy customer overall.

couldn’t the frontier labs simply compile their existing source code training data and have an assembly corpus to train on?

how is “yemen” not a word??

yes it may be valid for large values of “pretty soon”. the market is going to be irrational for as long as it likes :)

your follow-up makes no difference to the main point under discussion : if you ask two different llms about a legal fact pattern and they independently reference the same law, then it is, in fact, more likely that the law is more relevant to the fact pattern than if only one llm mentioned it.

if two llms independently cite the same law given a certain factual legal question then it generally makes it more likely that the law cited does, in fact, apply to the fact pattern in question ; i think you’re using “right” in a moral sense irrelevant to the comment i was responding to.

i suspect many licensed civil and other engineers may be looking at the currant situation as the moment when software people stop being able to just call themselves “engineers “ because they can type source code into an editor just like drawing a plausible bridge on paper doesn’t qualify you as being able to actually build one.

actually my feeling on this is that pretty soon unless you’re in the top 1% of performers for a software task it will become difficult to hold a role at a frontier company.

I’m at the end of my career and got lucky to share your interests at the beginning of a specialist boom

in your place i’d focus on delivering big wins for end-users via multiple roles ; i personally believe the days of being a specialist in one type of stack (ex: “i’m a ui developer”) are over , in this next stage we’ll say “i build products for young people looking for educational opportunities “ ; ie you’ll beed to cut across use-cases and disciplines in a way that was atypical just a few years ago

I’ve seen passionate artisans waste way more time than focused, disciplined, mercenaries; i think it’s a situational choice

testing is the new coding

i don’t condone cheating but i also don’t blame anyone who follows a generally benign career path for its financial benefits.

surely getting the same answer multiple ways suggests it is nore likely in the probability space?