HNHacker News
TopNewBestAskShowJobs

spencerchubb

1,275 karma · joined December 3, 2021

https://twitter.com/spencer_chubb
submissionscomments
spencerchubb··on Show HN: After 2 months my project management software is finally finished
Project management software is a solution, not a problem. An example of a problem could be that freelancers are forgetting things, or don't have time to get their work done, or have trouble finding clients, etc
spencerchubb··on Getting 50% (SoTA) on Arc-AGI with GPT-4o
It could exist many times. People can fork and clone the repo. People are likely to copy the examples and share them online.
spencerchubb··on Getting 50% (SoTA) on Arc-AGI with GPT-4o
Possibly

I think we need those pieces, and also a piece for determining hypotheses in an efficient manner. Monte Carlo Tree Search could be that piece. Probabilistically choose a node to search, and then backpropagate the probabilities back to the root node.

spencerchubb··on FTC sues Adobe for hiding fees and inhibiting cancellations
Somewhat reminds me of netflix's policy on sharing accounts. The CEO used to straight up say they don't care, and it's not really feasible to enforce account-sharing rules. Fast forward to today, and they "figured out" how to enforce it.
spencerchubb··on Bouba/kiki effect
I wonder if the ability diminishes in adults, or if adults are more inhibited and worried about looking foolish. Try the same experiment with adults after drinking alcohol!
spencerchubb··on SQLite is likely used more than all other database engines combined
OSS does not mean anyone can contribute. It means anyone can view source
spencerchubb··on AI Search: The Bitter-Er Lesson
TikTok is the digital version of this
spencerchubb··on AI Search: The Bitter-Er Lesson
The branching factor for chess is about 35.

For token generation, the branching factor depends on the tokenizer, but 32,000 is a common number.

Will search be as effective for LLMs when there are so many more possible branches?

spencerchubb··on Lynn Conway has died
Incredible

"The fact that I started a new career all over again, at the bottom of the ladder, after being fired by IBM and rejected by family and friends . . . may also give hope to others trapped in similar situations."

spencerchubb··on 23words.com
I like this concept. Wordle is very fun but I never liked how it was constrained to five letters. It felt like my brain was becoming very fluent in 5 letter words but not other words
spencerchubb··on A ChatGPT mistake cost us $10k
Yes Siri and Alexa can tell you the date. But I feel like you are assessing chatgpt unfairly. Would you say that a mobile phone is bad because you cannot run powerful programs like on a laptop? No, the mobile phone has its own purpose, which is convenience

Chatgpt can do many things that siri and alexa cannot: draft emails, write code, write unit tests, help me brainstorm, give me cooking recipes, or even talk about relationship issues. Every tool has strengths and weaknesses

spencerchubb··on A ChatGPT mistake cost us $10k
That is indeed a weakness of chatgpt - it's not so good with super current information

That is because it is only trained on info up to a certain date

It's good to know what tools are good and bad at

spencerchubb··on A ChatGPT mistake cost us $10k
A large language model is trained on vast amounts of text to predict the next token. That tool will have no idea what the current date is, unless the developers augment it by telling it the current date.
spencerchubb··on A ChatGPT mistake cost us $10k
I get the sense that you don't understand the purpose or design of chatgpt

No tool can do everything

spencerchubb··on A ChatGPT mistake cost us $10k
I agree this is more of a monitoring mistake, and little to do with chatgpt
spencerchubb··on Google Mesop: Build web apps in Python
I'm struggling to understand this demo app. You type in some text and it echoes the exact same text back to you? I assume the project just hasn't fully fleshed out the readme
spencerchubb··on Viagra improves brain blood flow and could help to prevent dementia
My username is my legal first and last name lmao
spencerchubb··on Viagra improves brain blood flow and could help to prevent dementia
Maybe, but maybe not. If I told you I'm either giving you actual viagra or a placebo, you might get more aroused even if given placebo. Because psychologically you'll be thinking "There's at least a chance I took viagra right now"
spencerchubb··on Viagra improves brain blood flow and could help to prevent dementia
Boss: You look... particularly excited to be working today

Me: I'm just trying this new technique to boost productivity

spencerchubb··on Mitsubishi robot solves Rubik's Cube in 0.305s
I did not intend to contradict but add on. I suppose my usage of 'actually' could be seen as contradicting
spencerchubb··on Mitsubishi robot solves Rubik's Cube in 0.305s
The fastest human solver (Max Park) actually has autism and problems with fine motor skills. Just goes to show that cubing is more about macro motor skills
spencerchubb··on Mitsubishi robot solves Rubik's Cube in 0.305s
The full solution seemed to be 14 moves by the axial turn metric, or 16 moves by the half turn metric. Axial means if you do turns on the same axis, it only counts as one move

This robot is able to do optimal solutions that a human wouldn't be able to find

spencerchubb··on Mitsubishi robot solves Rubik's Cube in 0.305s
Even if a human solver could start/stop with cube in hand, that would only save about 0.3 seconds
spencerchubb··on Mitsubishi robot solves Rubik's Cube in 0.305s
I'm a competitive speedcuber and my best time is a little over 5 seconds

It looks like this robot can do about 67 turns per second (tps)

The fastest humans can do 20 to 30 tps, but only for especially ergonomic algorithms. This robot was able to achieve its tps with arbitrary moves that would be terrible ergonomically for a human. Quite impressive

spencerchubb··on Ask HN: What was your most humbling learning moment?
I agree with this analysis. Another important factor is: How likely will this code actually be used and drive business value?

If very likely, then you should invest in making the code high-quality. If unlikely, then you should half-ass the code strategically

spencerchubb··on Better RAG Results with Reciprocal Rank Fusion and Hybrid Search
I'm curious, in your benchmark, what's the difference between BM25+Embedding and Embedding+BM25? And what do you use to make the embedding

If you make the embedding with an LLM, it should work for any language the LLM is trained on.

spencerchubb··on Better RAG Results with Reciprocal Rank Fusion and Hybrid Search
Except now the semantic capabilities are so much stronger. The transformer allows the model to get meaning from words that are far apart from each other
spencerchubb··on Transformers Can Do Arithmetic with the Right Embeddings
They do that in the paper
spencerchubb··on Transformers Can Do Arithmetic with the Right Embeddings
They do that in the paper. Least significant digit on the left
spencerchubb··on Ask HN: What is your ChatGPT customization prompt?
https://arxiv.org/abs/2404.15758
← PreviousPage 8 of 16Next →