Hacker News Reader

FWIW this (approximately) is what everybody (approximately) is trying to do.

killthebuddhaon Ogma: Interpretable Symbolic General Problem-Solving Model

On the one hand this is interesting to me because it's closely aligned with my own ideas about what will be the general form of the first AGI systems.

On the other hand, just like with my own ideas, there's very little concrete evidence that they're correct. It looks like the author has been very careful about wording and avoids saying directly "there exists an implementation of these ideas that seems to be working well enough to expect it to scale".

killthebuddhaon New findings point to an Earth-like environment on ancient Mars

What you're describing is one possible resolution and part of what's interesting. A very big open question is "which, if any, of the known factors are wildly rarer than we think?". Another possible resolution is that most of our estimates are actually ok and there's some other variable that's going to whack us on the head at some point.

killthebuddhaon New findings point to an Earth-like environment on ancient Mars

> To me it seems like the probability could have been anything.

The point is that we do know some things about the universe and, given what we do know, we think the probability is high. So we're obviously missing something but we don't know what.

The thing that makes the Fermi Paradox so interesting is that every resolution is either interesting or implies something interesting.

killthebuddhaon Theory of Constraints

A kind of radical belief that I hold is that basically everyone would benefit from working as a planner for a year or two, especially if you can spend a good amount of time away from your desk and out on the floor. The lessons you can learn by trying to coordinate a production floor are simultaneously very general, very practical, and oddly difficult to learn elsewhere.

killthebuddhaon Program synthesis

+1 to this. IMO the first versions of AGI are going to be program synthesizers.

killthebuddhaon You can't build a moat with AI

You could bootstrap a niche model using what's available and go from there. I don't think it's all that different from any other kind of bootstrapping startups tend to require.

killthebuddhaon You can't build a moat with AI

I don't see it that way. If you build a genuinely novel application then the critical data doesn't exist yet (IMO almost by definition). Sometimes it's easier for a startup to do this rather than for an incumbent who's trying to shoehorn the application into a preexisting framework (technical, operational, whatever).

killthebuddhaon The Texas Triangle: A rising megaregion unlike all others (2021)

FWIW they provide this [1] map. Dallas, Austin, and San Antonio are roughly collinear.

[1] https://kinder.rice.edu/sites/g/files/bxs4896/files/images/U...

killthebuddhaon Ethereum has blobs. Where do we go from here?

I think it would be useful if the full domain (vitalik.eth.limo) was displayed.

Not sure if that's possible or if it violates any HN policies about how links are displayed, apologies if it's a silly/useless suggestion.

Edit: Not sure how popularis eth.limo w.r.t. to HN submissions, but the full domain should probably be displayed for any eth.limo submission.

killthebuddhaon Barcelona’s Sagrada Familia will be completed in 2026

As another comment suggested, I think it's the crowds and the commercialism. It might as well have been erected inside Disney World. Maybe I'm insensitive, but it takes more than the literal/physical art (the building itself) to move me.

Edit: For example, just walking around the Gothic Quarter at night was a significantly more moving, cathartic, or "timeless" experience for me.

killthebuddhaon 2K earthquakes in 1 day off Canada coast

Wow, that playlist is absolute gold. These kinds of instructional, high-quality, long-form lecture series are IMO maybe the best thing on the internet. The kind of thing when I encounter it makes me take a step back and really appreciate the fundamental beauty and utility of the internet.

A few other examples:

  - https://www.youtube.com/playlist?list=PL9GwT4_YRZdBf9nIUHs0zjrnUVl-KBNSM
  - https://www.youtube.com/watch?v=ncd6q9uIEdw&list=PLND1JCRq8Vuh3f0P5qjrSdb5eC1ZfZwWJ
  - https://www.youtube.com/watch?v=ycJEoqmQvwg&list=PLbN57C5Zdl6j_qJA-pARJnKsmROzPnO9V
  - https://www.youtube.com/watch?v=Ps8jOj7diA0&list=PL9D558D49CA734A02

Sometimes I wish there was a frontend for YouTube that only has these kind of long-form lecture series, but I've never found one. I think part of the reason why is that essential "quality" is somewhat ineffable.

killthebuddhaon Fig is sunsetting

Not a direct answer to your question but

> We have several exciting new CodeWhisperer for command line features coming in 2024 including support for Linux

killthebuddhaon Ask HN: Which skills are least likely to be replaced by AI?

Spoiler: this comment does not technically answer OP's question.

Many of the answers in this thread talk about skills where AI seems unlikely to have any direct sort of impact at all (like plumbing). In my opinion, that could easily end up being technically correct but exactly wrong in spirit (assuming the ultimate goal is to have a good job). It's somewhat likely that AI-augmented or AI-enabled professions become so ludicrously profitable that any other profession becomes barely livable.

Instead of trying to avoid AI it might make more sense to ride the wave. Dive right into some industry that's undergoing upheaval, work for AI-powered startups, learn about the technology, and meet the future head-on.

I don't want to sound too "gung ho", I'm a software engineer who feels like my profession will be radically different in less than 10 years and I'm genuinely worried about staying relevant. But I think "staying relevant" is probably a better strategy than trying to get out of way.

killthebuddhaon Our next-generation model: Gemini 1.5

10M tokens is an absolute game changer, especially if there's no noticeable decay in quality with prompt size. We're going to see things like entire domain specific languages embedded in prompts. IMO people will start thinking of the prompt itself as a sort of runtime rather than a static input.

Back when OpenAI still supported raw text completion with text-davinci-003 I spent some time experimenting with tiny prompt-embedded DSLs. The results were very, very, interesting IMO. In a lot of ways, text-davinci-003 with embedded functions still feels to me like the "smartest" language model I've ever interacted with.

I'm not sure how close we are to "superintelligence" but for baseline general intelligence we very well could have already made the prerequisite technological breakthroughs.

killthebuddhaon Hi everyone yes, I left OpenAI yesterday

FWIW if you hold down the big white button it won't submit until you release it. I had no idea this was a thing until seeing someone tweet about it.

killthebuddhaon What if I’m wrong? (2023)

Hmm. I always thought "not even wrong" meant "correct, but answering the wrong question" or more generally "correct but irrelevantly so".

killthebuddhaon A distributed systems reading list

FWIW I don't see mention of incompleteness on the book's site http://book.dtornow.com/

killthebuddhaon ElixirNitpicks

Genuine question: From an application developer's perspective, what's the difference between mutating a value and transparently rebinding an old name to a new value? Is it just that in the latter case other references don't pick up the changes? So with rebinding we don't have something like

  a = 10
  b = a
  a = 11
  print(b) // 11

?

killthebuddhaon ElixirNitpicks

FWIW this is the #1 best use case I've found for ChatGPT. I recently started learning golang and as an experiment decided to use nothing but ChatGPT for reference/documentation. So far, ChatGPT has satisfied roughly 100% of my questions. A lot of the questions were like "what's the go equivalent of X in JS/TS?".

The returns seem to be diminishing after ~6 weeks, but the first 2-3 weeks were the least friction I've ever encountered learning basically any new topic. It was kind of wild, I don't remember getting stuck even a single time, it was just super high-density progress, non-stop.

More generally, the situation where "I have a good amount of generic knowledge, so I know what I want, but I don't have the specific knowledge I need" seems to be a sweet spot for ChatGPT.

killthebuddhaon Goodbye non-KISS appliances

Related, thinking out loud:

Why is the following timeline counterfactual (as far as I know) for basically all household appliances:

  1. it was invented
  2. it was commercialized
  3. some decades of varying modes of specialization, advancement, innovation
  4. reach technological maturity/saturation
  5. open source designs with idiot-proof manuals emerge
  6. everyone uses the open source versions
  7. maybe a freelancer market emerges
  8. it's no longer commercially viable
  9. it becomes a near-zero-cost improvement on the baseline human condition
  10. focus on new technology

The hopeful view (IMO) is that we're just at a point historically where things like washing machines and refrigerators are still at step 4 and people like me are starting to wonder why we're not further down the road? In 10-20 years maybe we'll be at 6 or 7?

The cynical view (IMO) is that the capitalist system operates by pulling a bait and switch on the average consumer. As technology becomes commercially viable, we use the slack it creates to pump everyone full of skittles and pretty little liars and then sell refrigerator-television-vending-machine appliances until the Decadent Society collapses under its own weight.

Maybe a more moderate view (IMO) is that it's naive to think that everyone could possibly have the time or desire to maintain their own refrigerators? FWIW My take on this particular view is that we should be trying to develop a society where it's not naive, where the average human stands on top of centuries of innovation rather than cocooned inside of it.

/idk

killthebuddhaon Why are we templating YAML? (2019)

I would love for someone to eviscerate the following idea:

Every deployed process receives exactly 3 environment variables, NONCE, TAGS and CONFIG_DB_PARAMS. Every process is bootstrapped in the same way:

  1. Initialize config db client.
  2. config = db.fetchConfig(tags)
  3. Use NONCE to signal config changes if needed.

Of course there's some environments where this Just Won't Work. I'm wondering if there are some very serious issues with this approach in a "standard" web application environment. It seems so straightforward but I've literally never seen it done before, so I feel like I'm missing something.

killthebuddhaon Before I reach my enemy, bring me some heads

I'd never heard of Sam Kriss before, but the writing drew me in and I spent a few minutes looking through his stuff. I have no clue about the veracity of the things he writes, but boy oh boy does he do the writing part well.

killthebuddhaon A Writer's Ruby

Is that his point? Or is his point that shipping rails with his own personal linting preferences is a good idea? (For what it's worth, I think it's a perfectly fine idea).

killthebuddhaon A Writer's Ruby

This post is so wonderfully confused it's throwing my brain into an infinite loop. Misunderstanding newspeak, then a howler of a non sequitur, all about elegant composition, all the while turning a simple announcement into a super clunky argument. It's perfect doublespeak, the kind of language that throws critical thinking into infinite loops.

killthebuddhaon People like me are why you shouldn't run a hosting company

In case the author is here: In the first code block we have `<KeyInsight ... />`, should it be `<Improvements ... />`?

killthebuddhaon Shane MacGowan has died

  One summer evening drunk to hell
  I stood there nearly lifeless
  An old man in the corner sang
  Where the water lilies grow
  And on the jukebox johnny sang
  About a thing called love
  And its how are you kid and whats your name
  And how would you bloody know?
  In blood and death neath a screaming sky
  I lay down on the ground
  And the arms and legs of other men
  Were scattered all around
  Some cursed, some prayed, some prayed then cursed
  Then prayed and bled some more
  And the only thing that I could see
  Was a pair of brown eyes that was looking at me
  But when we got back, labeled parts one to three
  There was no pair of brown eyes waiting for me
  
  And a rovin, a rovin, a rovin I'll go
  For a pair of brown eyes
  
  I looked at him he looked at me
  All I could do was hate him
  While Ray and Philomena sang
  Of my elusive dream
  I saw the streams, the rolling hills
  Where his brown eyes were waiting
  And I thought about a pair of brown eyes
  That waited once for me
  So drunk to hell I left the place
  Sometimes crawling sometimes walking
  A hungry sound came across the breeze
  So I gave the walls a talking
  And I heard the sounds of long ago
  From the old canal
  And the birds were whistling in the trees
  Where the wind was gently laughing
  
  And a rovin, a rovin, a rovin I'll go
  A rovin, a rovin, a rovin I'll go
  And a rovin, a rovin, a rovin I'll go
  For a pair of brown eyes
  For a pair of brown eyes

  And a rovin, a rovin, a rovin I'll go
  And a rovin, a rovin, a rovin I'll go
  And a rovin, a rovin, a rovin I'll go
  For a pair of brown eyes
  For a pair of brown eyes

killthebuddhaon US consumer spending dashboard built on data from 50M+ cards now on Snowflake

I'm interested in taking this approach to learning, would you mind sharing some good starting points?

killthebuddhaon Gmail, Yahoo announce new 2024 authentication requirements for bulk senders

If I wanted to learn everything there is to know about email and SMS spam/abuse policies, technical best practices, important standards, etc, what would be a good strategy? It feels like a super important but ridiculously intractable subject.

killthebuddhaon M2 Ultra can run 128 streams of Llama 2 7B in parallel

He founded a startup https://ggml.ai/