HNHacker News
TopNewBestAskShowJobs

Yenrabbit

863 karma · joined January 7, 2015

submissionscomments
Yenrabbit··on Boxed – Things I learned after lying in an MRI machine for 30 hours
And very quickly hit a rate limit apparently! Just switched to a different option without the unsplash limits.
Yenrabbit··on Boxed – Things I learned after lying in an MRI machine for 30 hours
"Every four seconds or so, you see something new that you would never have guessed from the previous pictures. Each time it's a different cascade of activations in your brain, evoking random memories, creating unexpected connections, and stimulating thoughts that would never have occurred to you...What I need is an app that does nothing but show you truly random pictures, with no curation and no memetic aspirations. If you know of one, please let me know." I made a site quickly, that shows a random series of pictures from Unsplash, since I was curious to try this! https://random-photo-cascade.replit.app/ (Built with Replit's new agent feature, which is impressive for this kind of very constrained task)
Yenrabbit··on The secret inside One Million Checkboxes
Fantastic telling of it in both text and video form. Great to celebrate these people doing the kinds of things we learned so much from! Thanks for sharing.
Yenrabbit··on How to build a 50k ton forging press
For a program started in the 50s, it's impressive that "Six of the ten presses are still operational today."!
Yenrabbit··on Show HN: The First Non-Smart AI Pendant (NotFriend)
Hey, I spent multiple minutes designing that! ;)
Yenrabbit··on Show HN: The First Non-Smart AI Pendant (NotFriend)
Thank you :D
Yenrabbit··on Show HN: The First Non-Smart AI Pendant (NotFriend)
Oh fantastic, glad others had the same idea :D
Yenrabbit··on OpenUI let's you describe UI using your imagination
Neat project! Also fun to see it's Chris Van Pelt (co-founder) doing most of the hacking on this.

In my first few tests, it was pretty meh (not as good as Vercel's V0) but then I figured out how to change the model away from GPT 3.5 and it got a lot better. I also appreciate making the system prompt etc. configurable!

Suggestion: Using placeholder images whenever an image/icon is included would be nice.

Yenrabbit··on SpaceX Raptor 3, SN1
Interesting how much less "stuff" hanging off it there is, I imagine the incentives are to add complexity so removing it over time is impressive
Yenrabbit··on FastHTML – Modern web applications in pure Python
Which example? I see global vars in a couple of the demos. The game state for the Game of Life makes sense, since it is intended to be shared. The `messages` list in the Chatbot demo is definitely NOT how you'd build a multi-user application, that's mainly showing the styling aspect.

In general, you'd have an actual database and make it so users can only see their own data! See https://github.com/AnswerDotAI/fasthtml/blob/main/examples/a... which adds a filter to queries and DDL statements to ensure that the user can only see/edit their own todos.

Yenrabbit··on FastHTML – Modern web applications in pure Python
I've been using this while it was in development and it's a pleasure to work with. Highlights for me:

- Incremental complexity - starts super simple and I can add stuff as I need it. I don't like frameworks where step 1 already leaves you with lots of files and a bunch of things you need to know.

- Easy escape hatches. I like some of the simpler demo/dashboard things but inevitably hit a ceiling that suddenly requires a lot of hacking to get past. Since FastHTML is a lot more transparent it's very easy to go right in and do something with JS or mess with the request or make something custom. So you're not stuck with only the widgets a framework gives you or anything like that.

Yenrabbit··on I am starting an AI+Education company
It's optimistic, but given the OP is one of the best-informed technical generative AI researchers, and has been passing on that knowledge to the rest of us for a decade +, I don't think we can just dismiss it as unfounded hype :)
Yenrabbit··on Bouba/kiki effect
These have entered our common household vocabulary after I called one of our houseplants "a little too kiki for my liking" and was forced to explain. Humans love categories, it's great to have names for splits like this :)
Yenrabbit··on Kino: Pro Video Camera
Delicious is the right word, everything (post included) is crafted with care.
Yenrabbit··on The Unusual Evolutionary Journey of the Baobab Tree
There is some contention around this. For a while it seemed like there could be two species on mainland Africa (https://en.m.wikipedia.org/wiki/Adansonia_kilima) and while most happily accepted the return of the status quo with the paper that decided they were all the same, I'm not entirely convinced (source: I collected location and morphology info from lots of these trees a decade ago and a number of traits were very distinctly bi-modal, although this is of course not proof that they're actually separate species). The Australian author who described A. kilima thought it was the diploid ancestor of the Australian species, rather than A. digitata or any of the Madagascan species...
Yenrabbit··on The Eric Ries Show: Interview with Sami Inkinen [video]
"He Sold His First Company For $3.5 Billion, Now He’s Reversing Type 2 Diabetes For 100 Million"
Yenrabbit··on Pydantic Logfire
They also have a UI framework (fastui). They have funding and devs looking to be useful, I think it's fine that they get to try out new things that work well with their core product.
Yenrabbit··on GitHub Copilot Workspace: Technical Preview
Despite being widely repeated they've publically denied this claim a few times (for e.g. https://x.com/natfriedman/status/1712140497127342404, https://x.com/simonw/status/1712165081327480892)
Yenrabbit··on GitHub Copilot Workspace: Technical Preview
Looks extremely cool. Have you shared / do you plan to share examples of this in practice? I would be great to read some case studies or even just browse some actual PRs rather than trying to judge from the catchy video :)
Yenrabbit··on Efficient finetuning of Llama 3 with FSDP QDoRA
This is exciting because:

- "Parameter Efficient" finetuning methods let you customize LLMs without having to train all the parameters

- But LoRA (the most popular method) didn't match full finetuning performance on some tasks

- DoRA closed the gap while still being very efficient

- Quantization (representing the original weights with fewer bits per parameter) makes things even more memory-efficient

- FSDP lets you spread the work over multiple GPUs, using less memory on each one.

The upshot is that where you previously needed, say, 8 fancy Nvidia A100s to fine-tune an LLM you can now do so on a few 3090s, and while it might take a little longer you're at least getting something almost as good as (or in some cases possible better than) the full finetuning equivalent.

Yenrabbit··on Ask HN: What rabbit hole(s) did you dive into recently?
Fun exercise to try and list them out! My last couple of weeks:

- 3D-printable parts storage solutions (via: I found some part storage bins in the discard pile at a local hackerspace)

- MITM proxy to snoop on Github Copilot API requests (via: we're building an jupyter AI assistant thing and got curious how other players do it).

- DIY robot arms (via: I'm making several for a nested 'you pass butter' joke, via a casual conversation about robotics being accessible now. YouTube is amazing at surfacing smaller makers once you start watching a few videos on a given topic)

- Learning about Oauth and JWT (via: 'why is auth still a pain?')

- Invertebrate UV fluorescence (via: that millipede is glowing under my UV torch!)

(a small subset of these end up documented https://johnowhitaker.dev/all.html eventually if you're curious to see a longer historical list)

I like rabbit holes where following the curiosity gradient to a satisfying conclusion is possible. "How does X work" leads eventually to code that does X. I'm less happy when they lead into a tangle of complexity, like digging into a library only to find weird abstractions 6 layers deep or trying to compare 18 different alternatives in a field I don't know very well.

OP I'd also like to hear yours!

Yenrabbit··on The rise and impending fall of the dental cavity
"The average person has enough of these bacteria in their gut to have a natural blood alcohol level - even after zero drinks - of about 0.1 mg/dl. Under pessimistic assumptions, BCS3-L1 will add another 0.2 mg/dl, bringing the total to 0.3. This is still a pretty normal number that some people have naturally (it would bring the average customer from the ~50th to the ~80th percentile of natural blood alcohol). It’s also far from the usual threshold for feeling tipsy (30 mg/dl) or too drunk to drive (80 mg/dl).

Under more realistic assumptions, the amount of alcohol produced by BCS3-L1 probably isn’t significant even by the very low standards of natural blood alcohol concentrations."

(From https://www.astralcodexten.com/p/defying-cavity-lantern-biow...)

Yenrabbit··on Homemade 6 GHz pulse compression radar
Fantastic work! I enjoyed reading the previous posts as well, very interesting.
Yenrabbit··on Low Cost Robot Arm
Haha fair play, this is exactly how things should be!
Yenrabbit··on OpenCat: Open-source robotic pet quadruped framework
It's slow, palm-sized and worse than a normal RC car in pretty much every way. This is not the mobile gunning platform you're worried about
Yenrabbit··on Fixing Gemma Bugs
Incredible work by the author stepping through all the nitty-gritty details and showing how easy it is to miss something subtle that could degrade performance.
Yenrabbit··on Yi: Open Foundation Models by 01.AI
Ooh I also use connections as a benchmark! It tends to favour things with 'chain of thought' style reasoning in the training mix somewhere since directly producing the answer is hard. Do you have public code you could share?
Yenrabbit··on Fine tune a 70B language model at home
We've got a technical post with all the juicy details coming next week. But that bit refers to packing the 4-bit weights into a type FSDP is happy to shard (like float16 or float32) which matches the other non-quantized bits of the model. This way FSDP will happily wrap and shard all the parameters as if they were just normal floats.
Yenrabbit··on Building a deep learning rig
Note that they shared part two recently: https://samsja.github.io/blogs/rig/part_2/

For those talking about breakeven points and cheap cloud compute, you need to factor in the mental difference it makes running a test locally (which feels free) vs setting up a server and knowing you're paying per hour it's running. Even if the cost is low, I do different kinds of experiments knowing I'm not 'wasting money' every minute the GPU sits idle. Once something is working, then sure scaling up on cheap cloud compute makes sense. But it's really, really nice having local compute to get to that state.

Yenrabbit··on Show HN: I made a Pinterest clone using SigLIP image embeddings
Neat! How many images are in the dataset out of curiosity?
← PreviousPage 2 of 7Next →