pstorm

Cost and intelligence aren't the only axes. Terra has better latency and output speed than Sol for example.

I’ve been building this out too, and your comment made me realize the missing piece for me. I’ve given the agents tools to validate its own work, but I haven’t improved the experience of humans verifying the agents’ work.

I played again and found issues. After level 40ish, basically nothing can hurt you. You heal too fast. And once I got past level 80ish, I was running so fast, I ran through walls. And the saddest part, I got to level 200+ and couldn’t die to lock in my high score! It was a fun time though, thanks for sharing this game!

This was more fun than I was expecting. Nice way to waste 20 min!

At a minimum, you increase top-k to cast a wider net, then after reranking, take the N you really want. You have to play around with it a bit, but that’s the idea.

I’m very surprised this isn’t getting more attention. Am I missing something?

It seems at or above SOTA on the given benchmarks, doesn’t have context rot, is orders of magnitude faster, and uses less compute that current transformer models. I suppose it’s just an announcement and we can’t test it ourselves yet.

Just fyi, for RAG/similarity search, adding a reranker was much bigger pay off than switching embedding models.

I built one of the top 3 results on Google when you search “compound interest calculator” and a dozen other similarly popular calculator pages.

The value isn’t the interface, it’s the trust that its calculations are accurate. I can’t tell you how many meetings I had with accountants and finance people to validate all the calculations.

Being flexible is important, but the latest research shows that heavy lifting improves flexibility about as much as a dedicated stretching program: https://pmc.ncbi.nlm.nih.gov/articles/PMC9935664/

That’s a new fire as of a couple hours ago - Sunset Fire. There are 5 going on in LA at the moment.

Less than 10% of my projects ever made anything. Check out indiehackers.com - tons of old posts of people starting things, but the domain is dead when you click through.

I’ve worked in e-commerce for years and the thing that always slows down the sites the most is 3rd party scripts. Are you addressing this? I couldn’t find anything in the repo.

Ive had websites slow down 10x just by introducing the Facebook re-targeting script for instance

Fyi regarding cover images: I have built and run a handful of book related websites and Amazon is the easiest place I found to get book covers. You just need the Amazon id and every image is a standard url.

This resonates a lot. For years, I would dive in for a week or two, sometimes I would make it to a few months. Lately, I've just been telling myself no, to keep focus on my start up, but it sometimes feels like I'm squandering this spark of energy you described. I absolutely love your approach of doing the planning in that initial burst rather than just getting some part of the idea working in a day or two.

I think those ultra-productive days earlier in my career account for some of the best learning, but now, over a decade into software engineering, focusing on the planning aspect seems much more prudent.

I've recently come to believe the differences in hypertrophy are negligible between eccentric and concentric focused movements, but can't find any recent, compelling research that says so. There was a 2017 meta analysis by Brad Schoenfeld (basically THE hypertrophy researcher) that showed a pretty significant different in hypertrophy: an average of 10% vs 6.8%.

I know Greg Nuckols from StrongerByScience believes this is mostly caused by lifters, especially untrained lifters which most of the research is on, having spent less time in eccentric phases so there is more opportunity for growth there, but it will eventually plateau.

This is RAG. They are retrieving specific info to augment the generation

I'm trying to understand this approach. Maybe I am expecting too much out of this basic approach, but how does this create a similarity between words with indices close to each other? Wouldn't it just be a popularity contest - the more common words have higher indices and vice versa? For instance, "king" and "prince" wouldn't necessarily have similar indices, but they are semantically very similar.

Came here to say this exact thing. I have the exact modem/router mentioned in the article - my reliability, speed, and coverage all improved when I switched to my own $100 router.

I looked into this to see where it was getting new information, and as far as I can tell, it is searching wikipedia exclusively. Useful for sure, but not exactly what I was expecting based on the title.

I’ve never really thought about this before, but it is obvious in hindsight. When I finally made enough money to move into a newer apartment, the noise from neighbors and outside was almost nonexistent. It was a stark contrast, but I never thought about how that affected my sleep/productivity, let alone how it would affect all apartment dwellers.

While I haven't read 3BP yet, you might be interested in the Zones of Thought series by Vernor Vinge. Extremely grand and unique universe, some politics, diverse cast, lots of interesting tech. I use it as a benchmark for grand space operas.

I'm surprised this isn't getting more love. I love the concept of finetuned, hyper-specific, tiny LLMs. Of course, the data is the most important part.

Based on #2, it seems like they only use the LLM when the page changes. I had a prototype of this sort of system working and it was surprisingly fault tolerant.

You raise a good point - AI + human review might end up being more time than just a human doing everything. I can see a certain subset of issues could be simple enough to done by AI and a quick review - like changing a button color or fixing clearly defined bugs. Time will tell how much work gets shifted over to AI + human review, but I'm betting on most of it.

What percent does an average junior engineer solve? If it is even close, these models can be run all day and night for cheaper than one yearly SWE salary.

I've been planning on building some of this for an internal tool, but now it looks like I don't have to. I'm impressed by the demo, it looks really polished.

I'm particularly surprised by the speed considering all of the pre and post processing. I am doing some similar things and that is one bottlenecks. I'll dig in, but I'm curious what models you are using for each of these steps.

500%? So if the generative AI model created something worth $1m, you tax it $5m? How do you tax a technology anyways?

Where/how do you host your finetuned small models? I have some use-cases, but that headache always leads me right back to OpenAI.

Sorry late reply. I believe the goal of separating by age is to have a more apples-to-apples comparison and to see trajectories. For instance, it highlights that at a young age, Gen Z is actually ahead of previous generations on homeownership, but also the upward slope looks like it may trend flatter. It isn't drastically different, in this case, than just looking at the averages.

A few theories:

- They have better just stuff, but aren't release it yet. They are at the top already, it makes sense they would hold their cards until someone got close.

- They are more focused on AGI, and not letting themselves get side tracked with the LLM race

- LLMs have peaked and they don't want to release only a minor improvement.