HNHacker News
TopNewBestAskShowJobs

ldhough

81 karma · joined February 17, 2023

submissionscomments
ldhough··on The EU AI act is coming, this time for real probably
> donate to https://noyb.eu/en

Thanks for the referral, joined as a supporting member :)

ldhough··on Scientists Use GPT AI to Passively Read People's Thoughts in Breakthrough
This was always one of my favorite shows. For (what started as) a police procedural, it was incredibly prescient in many ways.
ldhough··on Agile soccer skills for a bipedal robot with deep reinforcement learning
FRC?
ldhough··on Some employers have decided to build their own housing for workers
I did read that but it seemed pretty clear that the comment thread and the comment you were responding to were not in fact discussing one specific instance, but rather the broader phenomenon:

> these company investments ...

> If the companies ...

> Having your employer [not "having Cook"] as your landlord ...

Although given that your comment was clearly about the case study I should have made clearer that my intent was to communicate that just because Cook isn't renting the houses that the parent commenter's concerns are still valid because more companies are.

ldhough··on What is it like to be a bat? (1974) [pdf]
It seems to me like the answer to "would you learn anything new by being a bat?" is necessarily yes, because you would at least learn the answer to the question itself.
ldhough··on Some employers have decided to build their own housing for workers
I've seen people say this a few times here but the article says:

"a growing number of employers around the country have decided to build their own housing for workers, mostly for them to rent but sometimes to buy."

ldhough··on Whistleblowers are the conscience of society, yet suffer gravely
Bit of an ironic stance for someone with the name Galen Erso to take ;)
ldhough··on 1Password to Add Telemetry
Was referring to the app yeah, I don't use the website.
ldhough··on Intel posts largest loss in its history as sales plunge
> almost never need a fan

While I am also concerned about and need the extra performance, this point seems somewhat underappreciated in discussions about the ARM macbooks. Fan noise is so incredibly annoying and distracting and I think not having it improves productivity.

ldhough··on 1Password to Add Telemetry
> thin and non-resizable

I just checked and this works fine on macOS

ldhough··on An example of LLM prompting for programming
Sadly I don't think this is true for art:

https://restofworld.org/2023/ai-image-china-video-game-layof...

I really hope it doesn't end up being the same with code :|

ldhough··on We need a more sophisticated debate about AI
I've tried tic-tac-toe with it (GPT4, web platform - maybe results are different with the API and temp 0?) easily a dozen times using various ways to prompt it over the past few weeks, it only played well once so I'm assuming it was just luck.

I usually have it go first and start with X but have tried O. It usually makes no attempts to block 3 in a row, even if I tell it "try to win," "block me from winning," etc. Once it told me I won without having 3 in a row, many times it plays after I've already won and then tells me I've won, though usually it does manage to recognize a win condition. Today I tried asking it "do you know the optimal strategy?" and it explained it but claimed that it hadn't been using it to make the game "more fun for me" (honorable but I'd already told it to try to win) and asked if I wanted it to play optimally. It tried and ended up claiming we had a draw because neither of us achieved a win condition even though I'd won and it just played after the game was over.

Various strategies include asking it to draw ASCII, provide moves in symbol-location notation, ex: X5, asking it how to play, telling it to try to win, etc.

I do find it very odd that it is so poor at tic-tac-toe, it seems to even handle seemingly novel games better.

ldhough··on Pause Giant AI Experiments: An Open Letter
I assume you're correct, I'm on the GPT4 API waitlist so I'll be able to better test these things once I get API access.
ldhough··on Deepest fish ever caught on camera off Japan
I tried a ~1x1cm cube of it when I was in Iceland... that assessment is pretty spot-on.
ldhough··on Asahi Linux M1 GPU drivers can now run Windows games via Steam Proton
Oh I don't at all, I was contesting

> this weakens Apple's grip.

Because I don't think Asahi harms them.

ldhough··on Asahi Linux M1 GPU drivers can now run Windows games via Steam Proton
I don't think it harms them financially though. You still have to buy the macbook and I don't think they earn any significant amount of money from you running macOS on it after that, I've bought like one $6 app on the mac app store. Being able to run linux makes me more likely to buy more macbooks in the future and if I can run windows games I'll get the highest-specced one available.
ldhough··on Pause Giant AI Experiments: An Open Letter
I was just using the ChatGPT web UI (with GPT4) so I didn't manually set the system prompt or the temperature. I checked all of yours and they do exist with the right researchers. I tried again in the web UI and it generated junk, when I told it to only cite articles from 2020 and 2021 it got them mostly correct but invented one and put the wrong authors on several.
ldhough··on If AI scaling is to be shut down, let it be for a coherent reason
I mean then what is the basis for calling him not a "real practitioner" while Yann LeCun is? It seems a little unfair to dismiss outright the opinion on a deep learning technology from a guy who won a Turing award for research into deep learning. I also don't see any evidence that he is "transparently self-interested."
ldhough··on If AI scaling is to be shut down, let it be for a coherent reason
Yoshua Bengio is a signatory and shares a Turing award with Yann LeCun though...
ldhough··on If AI scaling is to be shut down, let it be for a coherent reason
I'm very much not an expert either but apparently for the regular "attention" mechanism memory and compute requirements scale quadratically with respect to input sequence length. So increasing it by just two orders of magnitude would mean (I think) a full context window needs 10000x more memory and compute time and presumably costs would go up by at least as much. GPT3 (I think) uses the regular attention mechanism while 4 is unknown.

However, GPT4 claims there are techniques to improve scaling (complexity down to sub-quadratic or linear) without affecting accuracy too much (I have no clue if true): sparse attention, long-range arena, reformer, and performer.

I'm also pretty sure I've read (and anecdotally it seems true) that accuracy decreases with longer input/output sequences regardless. How much I also don't know.

ldhough··on Pause Giant AI Experiments: An Open Letter
I tested the Base64 thing with GPT4 and it produces code that does seem to work. There have been other tasks I've given it (C++, Clojure, JS) that it doesn't get on the first try or in some cases doesn't get at all though. One task I tried in C++ it kept going in circles and ignoring requirements from prior prompts and I tried numerous ways to prompt it.

All that in mind, I'd be lying to say I'm not more than a little concerned with the progress from 3.5 -> 4. I'm only two years into my career and my fingers are crossed that it won't significantly impact the market for devs for as long as possible.

ldhough··on Pause Giant AI Experiments: An Open Letter
Portobello mushroom is certainly wrong, I also would have included color for salmon but that one is more up for debate I think. I think some humans would probably write the same categorizations though.

[[[Me]]]

> I will provide two lists of items inside brackets, separated by commas. I want you to categorize the entities in the first list into the categories in the second, keeping in mind that an entity in the first list may fit into multiple categories: [goldfish, shark, salmon, blue, tiger, cake, cyan, portobello mushroom, tree] [animal, fish, food, color, mammal, plant]

[[[GPT4]]]

> Here's the categorization of the entities in the first list into the categories in the second list:

    goldfish: [animal, fish, food]
    shark: [animal, fish, food]
    salmon: [animal, fish, food]
    blue: [color]
    tiger: [animal, mammal]
    cake: [food]
    cyan: [color]
    portobello mushroom: [food, plant]
    tree: [plant]
ldhough··on Pause Giant AI Experiments: An Open Letter
Fwiw they are right that it produces bogus citations depending on what you ask. I did their tree example with GPT4 and got the following:

------ PROMPT --------

Can you give me citations for recent research on trees?

----------------------

====== RESPONSE ======

Certainly! Here are some recent research articles on various aspects of trees, including their ecology, physiology, and response to climate change. Please note that these citations were published after my knowledge cutoff date in September 2021, so I cannot vouch for their accuracy or the impact of the research. However, they should give you a good starting point for your research:

    Anderson, T. E., Gonzalez, M. A., & Smith, K. L. (2022). Tree species diversity and its impact on forest ecosystem services. Journal of Ecology and the Environment, 33(2), 147-158.

    Baker, P. J., Oliver, R. J., & Dearing, L. E. (2023). Carbon sequestration potential of urban trees under climate change scenarios. Urban Forestry & Urban Greening, 68, 102139.

    Chen, X., Xu, L., & Zhu, H. (2022). Tree-ring-based reconstruction of drought variability and its potential drivers in subtropical China. Science of The Total Environment, 796, 148896.

    Ferreira, J. T., Cardoso, S. M., & Sousa, P. M. (2023). The role of tree genetics in adaptation and resilience to climate change: A review. Tree Physiology, 43(1), 1-15.

    Grant, N. C., Matthews, T. J., & Wilson, R. J. (2022). Assessing the vulnerability of tree species to climate change: A global meta-analysis. Global Ecology and Biogeography, 31(4), 623-637.

    Leonard, S. D., Smith, D. W., & Thomas, E. M. (2023). The impact of tree age on leaf photosynthesis and water-use efficiency in deciduous forests. Tree Physiology, 43(2), 243-255.

    Nguyen, H. T., Sato, T., & Yoshida, T. (2022). Exploring the effects of tree diversity on forest productivity and stability: A global meta-analysis. Forest Ecology and Management, 509, 119102.

    Patel, S., Sutherland, W. J., & Diaz, A. (2023). How do trees respond to extreme weather events? A global synthesis. Global Change Biology, 29(1), 219-234.
Please make sure to access these articles through proper channels, such as a university library or research institution, to ensure you are reading the most up-to-date and accurate versions.

================

I didn't check all of them but the ones I did are seemingly completely fabricated, which makes sense because for some reason it tried to produce research from after 2021 despite seemingly knowing and mentioning the date of its knowledge cutoff.

ldhough··on GPT-4 Designed a Programming Language
Although even in this example it does say some things that are simply wrong:

"Some programming languages like TypeScript and Swift use a colon for both variable type annotations and function return types."

This is incorrect for Swift, which uses the same arrows as "TenetLang" for function return types. Actually the first thing I thought looking at the example code was "looks swifty but not quite as well designed."

ldhough··on Build full “product skills” and you'll probably be fine
While I do think there is some threshold where increased productivity makes positions redundant, I don't think 2x would do it in most orgs. My current team easily has enough work for us all to be 2x more productive.
ldhough··on Build full “product skills” and you'll probably be fine
Two members of my family have now called me on the phone a few times to ask me to ask GPT something, after I convinced them of its value and after I had them make their own account. They are absolutely smart enough to ask it but are very much non-tech people.

There is definitely a tendency of some people to avoid tech at all costs if they can and trust in "tech people." Fine with me, I ask them about my people questions often enough.

ldhough··on Build full “product skills” and you'll probably be fine
I've also been considering the concept of "AI-amenable" programming languages recently, and I think it would be quite an interesting twist of history if lisps became once again the AI languages of choice because of their simplicity. Going from the language to write AIs in to the language that AIs write.

Fwiw, GPT doesn't seem to write Clojure as well as Python or JS or even C++, which is obviously much more syntactically complex. That said there is probably an order of magnitude more training data for all of those.

ldhough··on More students are turning away from college and toward apprenticeships
"Cannot be taught" is a pretty strong claim. I'm not an artist but it is my understanding that at least some forms of art (classical, sculpting) require at least some degree of instruction, and being in school is probably going to provide easier access to resources like models. A quick google search with artists I'm familiar with (Michelangelo, Caravaggio, da Vinci) confirms all were apprenticed to other artists.

Anecdotally I think my programming skills also benefited from a formal education, though there are without a doubt many self-taught developers who far exceed my skill.

ldhough··on More students are turning away from college and toward apprenticeships
I agree but according to the BLS employment numbers are as follows in the US:

Electrician: 650k

Plumber: 469k

Earthwork: Don't see an exact match

Welding: 428k

Carpentry: 668k

Software Devs, QA, & Testing: 1.62m

Programmers: 152k

Not sure how they differentiate between devs & programmers but even if we just take the 1.62m figure it is well over half the total employment in those trades. If software devs get 75-90% replaced (I don't think it'll be this bad and for my own sake as an early career dev I really hope not but I don't see it as impossible) I imagine most white collar jobs are coming with us. Will the trades pay as well when a ton of people are looking to reskill into something that still exists?

ldhough··on Testing GPT 4's code-writing capabilities with some real world problems
I mostly work in Clojure and I've noticed this as well, it is usually more efficient to just write it myself if I know what I want to do. However for a language or stack I don't understand well where I just need something quick and dirty it really shines... I had to work with a JS visualization library and got it to write what I needed over 10-15 iterations, much faster than if I actually had to learn the library. General programming knowledge sufficed for noticing things that didn't look right and since the output was visual that helped to sanity check the output as well.
← PreviousPage 2 of 3Next →