ww520

Here is where airship makes economical sense.

That's what a LLM based template engine looks like.

Motorola Moto G is a very capable phone on the cheap. It's even cheaper if you don't mind older models or used.

https://www.amazon.com/s?k=motorola+moto+g&qid=1757978993&rn...

You are right.

> The current hardware includes 16 microLEDs and 16 photodetectors, supporting a 16-variable state vector…

Fiber optic is a communication medium, not a computation device.

Have they heard about staging roll-outs and A/B testing or Multi-Armed Bandit testing?

I was driving from outside of Boston. Parking near subway was atrocious as well.

I paid over $50 USD for parking a bit over two hours in Cambridge Massachusetts when visiting MIT. Parking in the Boston area is atrocious.

I’m just visiting Toronto now. The Path is amazing. This is one of the entrances across from the Meridian Hall.

https://imgur.com/a/Y4S40JC

Is there a way to use Google Map to navigate the Path from one place to another?

Nope. If they didn't actively work against us, we would thank the lucky stars.

From the little I know of, language based provers like Lean allow mathematicians to specify the steps of a proof in a precise language format that can be “run” by a machine assisted prover to prove its validity. It automates the verification and makes proofs reproducible.

In the past when proofs were done by hands, little mistakes or little changes could lead to a complete failure of the proof. Mathematicians spent weeks or months to redo the steps and recheck every little detail.

Machine assisted prover basically raises the abstraction level for theorem proving. People don’t need sweat about little details and little changes in the proof steps.

Language based machine provers also enable wider corroboration as a problem can be subdivided into smaller problems and farmed out to different people to tackle, perhaps to different experts for different parts of the problem. Since everyone uses the same language, the machine can verify each part and the overall proof once the parts come together.

Mad respect. This is an incredible project to pull together all these technologies. The crown jewel of a search engine is its ranking algorithm. I'm not sure how LLM is being used in this regard in here.

One effective old technique for ranking is to capture the search-to-click relationship by real users. It's basically the training data by human mapping the search terms they entered to the links they clicked. With just a few of clicks, the ranking relevance goes way up.

May be feeding the data into a neural net would help ranking. It becomes a classification problem - given these terms, which links have higher probabilities being clicked. More people clicking on a link for a term would strengthening the weights.

Or the incentive aligned.

A language level or std library level poly-fill for the missing SIMD operations for the platforms would be great.

Excellent write up. I really appreciate the clear and detailed explanation of the algorithm.

The nice thing about Zig’s SIMD operation is the register size support is transparent. You can just declare a 64-byte vector as the Block, and Zig would use an AVX256 or two AVX2 (32-byte) registers behind the scene. All other SIMD operations on the type are transparently done with regard to the registers when compiled to the targeted platform.

Even using two AVX2 registers for 64 bytes of data is a win due to instruction pipelining. Most CPU have multiple SIMD registers and the two 32-byte data chunks are independent. CPU can run them in parallel.

The next optimization is to line the data up at 64 byte alignment, to match the L1/L2 cache line size. Memory access is slow in general.

Zig is great. I have fun writing in it. But there’re a few things bug me.

- Difficult to return a value from a block. Rust treats the value of the last expression of a block as the return value of the block. Have to jump through hoops in Zig to do it with label.

- Unable to chain optional checks, e.g. a?.b?.c. Support for monadic types would be great so general chaining operations are supported.

- Lack of lambda support. Function blocks are already supported in a number of places, i.e. the for-loop block and the catch block.

Wow. This is very impressive. It would be trippy if IE can run https://win32.run again.

These are excellent tips. I especially like the debugger integration in build.zig. I used to grep the cache directory to find the exe. The integration avoids all the extra steps.

It should be sold at a higher price. You perform a service by “unlocking” the camera.

Vibe coding can get very far. Years ago I interviewed with a company whose business was doing price modeling on retail products. It was couple guys, who were not developers, writing some Excel VBA scripts and Access VBA scripts. They got big customers. Their system won't scale and was not robust; it's basically a proof of concept. They got tens of millions in funding. They hired a seasoned team of developers to build the real product. The project failed. They fired the team, and was interviewing me to restart. Their POC was basically vibe coding and actually got their pretty far. I didn't join because they had burnt through quite a bit of their funding, they had built the sales and marketing orgs while they didn't have a real product, they had fired the entire engineering team where all the organization knowledge had lost and it spoke of some deep distrust between engineering and the rest of the company, and their offer was completely not compelling.

Yeah, the light propagation videos are just high on misleading theories.

Building a MCP SDK for Zig.

That’s why founding engineers are such a raw deal. They take just as much risk as the founders but much less payout. Also on the hook to do most of the work.

When business approach me asking for automating a business process, I usually ask them to do the process in manual steps and create check lists for it. The check lists force them to think it through and are the agreed upon steps to perform the process. After they do it by hand for a while and find out where the pain points are, we can then automate. Since the check lists are already in place, automation is really easy.

Checklist is for people who have a fair amount of experience doing the task. It's just a reminder to perform the steps. An experienced chef reading "pre-heat the pan on low-medium.." would know what those terms mean.

Wow. This is so good. I would never imagine using PID control for animation motion. Saving this.

Thanks for bringing new perspective into the discussion. Your Rust code is a good example. I did take Rust for inspiration.

The Zig version ‘Interface1.implBy(Impl2)’ is a rough equivalent to ‘impl Interface1 for Impl2’. It did it within the constraints of Zig.

Jove was my first editor on Unix. Emacs took up too much resource and was too slow back then.

The pipeline for LLM to MCP and to the app looks like,

  LLM -> chat app -> MCP client -> MCP server -> specific app (Blender)

The chat app doesn’t know how to talk to Blender. It knows about MCP and links in a client. Blender exposes its functionality via a MCP server. MCP connects the two.