HNHacker News
TopNewBestAskShowJobs

islewis

293 karma · joined April 6, 2023

submissionscomments
islewis··on Building Meta's GenAI infrastructure
I know we won't get it this from FB, but I'd be really interested to see how the relationship of compute power to engineering hours scales.

They mention custom building as much as they can. If FB magically has the option to 10x the compute power, would they need to re-engineer the whole stack? What about 100x? Is each of these re-writes just a re-write, or is it a whole order of magnitude more complex?

My technical understanding of what's under the hood of these clusters is pretty surface level- super curious if anyone with relevant experience has thoughts?

islewis··on How ChatGPT made me lazy
I don't think "lazy" is the word I would use, but I can see how some would see it like that. In my mind, GPT (and LLM's) are just the next layer of abstractions on top of a already massive stack of technical abstractions. For a SWE, this stack of abstractions crosses so many levels for a simple "hello world" to work:

1. Software (my code) 2. OS (Linux, kernels, etc) 3. Hardware (3090, 5090x, etc) 4. Electrical (Where is my energy coming from? how is it produced?)

Each of these levels could be broken into another 10 abstractions: On the software level, some people may understand how their compiler is working, but could they program in binary? What about understanding how their program interacts with memory? What about the kernel of where their software is deployed in the cloud? Do they know how their software is deployed in the cloud? Could they build the production server rack that their container is deployed on? Obviously this gets a bit ridiculous the further down you go- it's impossible to have knowledge about every part of what makes your code work.

I think that when people use terms like "lazy" or say that knowledge is being lost with abstractions like GPT, they ignore the massive list of abstractions that allow them to be productive.

I'd guess my thesis is that newer/GPT-aided engineers don't necessarily have less understanding, but their knowledge might just be shifted by one level up on the abstraction stack.

islewis··on YC: Requests for Startups
I doubt there's much thought behind it, and just the fact that each "Request" is written by a different person.
islewis··on Hi everyone yes, I left OpenAI yesterday
Given Karpathy's draw towards teaching/educational content, I've wondered where he falls on the spectrum between Sam Altman's interpretation of "Open" in OpenAI, and someone on the opposite end (like Musk).

I'd imagine if one was fully onboard with the AI/LLM commercialization train, there's no better spot than OpenAI right now.

islewis··on Bard is now Gemini, and we’re rolling out a mobile app and Gemini Advanced
> Messy branding is now par for the course at Apple. The iPad line alone is something out of a Dell catalog: iPad, iPad 10thgen, iPad Air, iPad Mini, iPad Pro.

a comment from another thread this morning- https://news.ycombinator.com/item?id=39300741

islewis··on Markov Chains are the Original Language Models
I resource I found invaluable in demystifying LLM's was Andrej Karpathy's Youtube series. In particular, his "Makemore" video built a basic LLM in the same vein as OP's.

It's long (2h), but well worth the time if anyone is curious about the technical workings of LLMs.

https://www.youtube.com/watch?v=PaCmpygFfXo&list=PLAqhIrjkxb...

islewis··on TSA introducing self-service screening technology in Las Vegas
> After getting spanked by DHS OIG in 2017ish

I'd love to read more about this, but a quick google search didn't bring up much. Any good sources?

islewis··on Titled Tuesday Cheating
Having a quick look at ChessMonitor, it looks like much of the analysis tools mirror what can be found on Lichess (unsure about Chess.com). Is the value providing the aggregate analysis between Lichess and Chess.com? Or is there functionality past what can be found on the respective sites?
islewis··on Serverless at Scale: Lessons from 200M Lambda Invocations
> why their calculations show that 6rps (that's what their "17 mln monthly lambda invocations" really means) would require 25 servers

I'm guessing that it has something to do with the average job taking 15 minutes. 6rps represents 6 jobs being created per second, but each one takes 15 minutes to run until completion. Another way to look at it is each second 90 minutes of lambda work is created.

If you consider 15 minutes the rolling window, which looks like a fair assumption based on the graph provided, there could be up to 5400 (15 min x 60 sec x 6 rps) functions running at once. Working backwards, 25 medium instance provide (25 instances x 4 GB mem) an 100gb memory pool, or 100,000mb. That leaves around ~18mb for each of those 5400 jobs, if you don't consider OS resource overhead.

Looking at averages in this situation very possibly can give a warped perception of reality, but 25 instances doesn't seem out of the realm of possibility. I'm sure they have much more relevant metrics to back up that number as well.

Whether the functions really need this much time to run is another issue entirely, and hard to answer with the information given.

← PreviousPage 3 of 3