steinsgate

We found something surprising about ARC AGI 2: the benchmark aiming to measure human-like fluid intelligence. Just enabling a stateful Python tool boosts performance across models. We got > 4x performance improvement in GPT OSS 120B (high). The effect continues well into frontier territory (GPT 5.2) with double digit gains.

We aren't sure whether these gains happen because code execution is a stronger form of verification compared to pure CoT or because it encourages qualitatively different thinking patterns.

Another interesting finding: interleaved thinking, the model capability behind these gains, seems fragile at the infra/client layer. Soft failures can make capable models look much worse than they actually are.

Just wanted to mention that there are indications that universal constants e.g. fine structure constant, gravitational constant etc. are time varying [0].

[0] https://en.wikipedia.org/wiki/Time-variation_of_fundamental_...

Ask HN threads contain a lot of valuable information. So I decided to categorize and organize this information in a GitHub repo. Would appreciate your feedback on the following:

1. Do you find this to be of any value?

2. Any suggestions on what you would like to see in such repositories?

Hi everyone. I recently created Ask HN summaries. It is a Medium publication that turns an Ask HN thread into a blog post and an accompanying GitHub repository. I would love to hear your feedback on it.

I have been working on a NLP project where I needed to identify different forms of the same word. Typically, this is done by Stemming and Lemmatization. These methods are not accurate, and I needed high accuracy in my project. Since I found no libraries/packages that can do this, I decided to write a Python package myself. It works quite well now. It is also trending in /r/python. Feel free to check it out, I would love to hear your feedback.

I have been working on a NLP project where I needed to identify different forms of the same word. Typically, this is done by Stemming and Lemmatization. These methods are not accurate, and I needed high accuracy in my project. Since I found no libraries/packages that can do this, I decided to write a Python package myself. It works quite well now. Feel free to check it out, I would love to hear your feedback.

If someone told me that there would come a day when two of the first three HN posts will be about Bob Dylan and Leonard Cohen, I would have dismissed them as high or delusional. Turns out that I would be very wrong.

Nice work! You said that you avoided machine learning because labeled data is hard to find. What about unsupervised approaches?

Frankly speaking, I am a bit skeptical about pattern matching algorithms for answering questions. It would help if you showed some kind of stats about your algorithm's performance on a diverse question set. For example, you can scrape simple quiz questions (and answers) from quiz sites [1] and report back on the performance.

[1] http://www.quiz-zone.co.uk/questionsbydifficulty/1/0/answers...

It is a free service as of now (at least for me, initially they said I would have to pay after an year, but this never happened). Technically, they don't even have a business model till now. What they have is scale. I think they are afraid that a subscription model will make users churn and they would end up losing the scale. The only other alternative is an ad supported free model or a freemium model. I am wondering if a freemium model would make sense for WhatsApp.

What's the value of WhatsApp? It is the scale of the product. I use it because most of my friends are on it. There are many other alternatives to WhatsApp as pointed out by many in this thread, but none of them have nearly the same traction.

The only business model that allows such scale is the ad supported model. If WhatsApp was subscription based, I am sure they could not have achieved this scale. And I wouldn't have found value in it.

I realize that the ad supported model is what indirectly adds so much value to WhatsApp. Therefore, it makes sense to me to support that model if the model is reasonable enough. They have end to end encryption, which means my content is safe from prying eyes. That's already huge. So I really don't mind if they share my number with Facebook.

another TL;DR : Making mistakes is a great way of learning. To learn from mistakes, the first step is to identify that you have made an error. The second step is to correct it. When we grow older, we become worse at the first step i.e. identifying the error. Without identification, there can be no correction and no learning.

This is one of my favorite posts from Paul Graham. I would add one more point to it based on my experience.

- Make something that has value from day one. Many tech based startups have significant value for users only when many other users are also using it. Tech based political startups, for example, suffer from this problem. These kind of startups are harder to build because getting and retaining the seed users can drain you easily. It is easier if you build something that adds value to the life of the very first user, even if he is the only one using it.

I am a big fan of Floyd and his solo stuff. He is not very difficult to see these days, he has been touring non stop for an year now. Maybe you can still catch him.

I was wondering if the show discovery space is crowded because of Songkick (which I already use), Bandsintown etc. What is your thought on that? On the other hand, Songkick at least has little to no social aspect, which kinda sucks.

Didn't know of this website. So damn useful!

I just registered as @darkvibration and put in all the concerts I have been to. Is this a purely personal journal kind of thing or have you been thinking of adding some social feature to it as well?

I would love to know your vision/plan with the site.

Love the caption. Straight out of the future!

Best of luck getting your sis to quit :-)

I was skeptical at first too. But man, was I wrong! I have recommended it to about four or five friends so far. Only one read it till the end. And it didn't work so well on him, even though he reduced smoking. Did the book work on your friends who read it completely?

Yes that's a great book. I am still amazed how well it worked for me. I was a chain smoker smoking 40 cigs a day, and then I read the book and poof! I stopped for good.

I heard that this is related to the arrest of the Kickass torrents admin. But I don't understand why a search engine would need to shut down.

Our final goal is to predict results using deep learning. Good luck to you too!

Yes, this is indeed training data. But you are absolutely right, we would definitely need to inspect and analyze the data for sanity checks. Therefore storing to a DB that has nice analysis features probably makes sense. Thanks for your answer!

Thanks, that analysis helped me a lot.

Thanks for making this app. I had a similar idea too, but never executed it. Of course, the idea was based on a certain need. Often, I would have unscheduled events such as "talk to Santosh sometime next week" or "go on a short bike trip with Francesca next month". Basically events that have not been assigned a defined date and time. These are the events that I would like to see in the week view or month view (when I zoom out) so that I know that I need to find a specific time for it. The other events (which have a set date and time) receive lower priority in the zoomed out views. Have you thought of implementing something similar? Would be nice.

Confession : I am guilty of using regex superpowers to extract data from urls. Will check out w3lib soon!

It is possible, but I think if you calculate the probability of the neutral mutations happening in series, this probability is astronomically small.

It is possible for all the mutations to appear simultaneously as well. But again, the probability would be astronomically small.

Thanks for looking into the original document and providing a detailed explanation. You are basically saying that the journalists sensationalized the findings. The researchers do not claim that they traced the most ancient mutation responsible for multicellularity. The researchers also do not claim that this single mutation created an almost functional multicellular organism. Am I getting this right?

I think you are talking about something known as hitchhiking mutations. Suppose an organism develops two mutations A and B, which lie close together in the chromosome. A is beneficial and B is neutral. When the chromosome is copied during reproduction, the copy will contain both A and B (because they are nearby). This copy will have an overall advantage because of the presence of A, and will spread in the population. https://en.wikipedia.org/wiki/Genetic_hitchhiking

One common objection to the theory of evolution claims that the theory is self contradictory. It says that it is impossible for complex features such as wings, eyes etc to evolve because they will require a series of mutations that do not individually confer any evolutionary advantage. Since they do not have any evolutionary advantage, they cannot spread in the population. Therefore the chance of these mutations appearing in series is almost as small as these mutations appearing simultaneously. The chance of these mutations appearing simultaneously is, of course, astronomically small.

The usual counterargument says that just one mutation is usually enough to create a functional prototype of a complex feature/organ (MVP, if you please). A series of mutations is not required.

This study provides yet another example backing this counterargument. One single mutation can change an unicellular entity and make it behave like a multi cellular organism. Intuitively, one would think this would take many mutations. However, intuition is wrong in this case, and in many others.

I think this is a perfect example of a pseudo scientific article. The author claims there is no free will based on another claim : "laws of Physics can describe/predict human actions". Physics does not make this second claim, at least, not yet.