mhast

It depends on the model.

It's typically ok for MoE models but if you try to run something non-MoE the speed will plummet. In that same thread there are people getting 50 tok/s on MoE models and 5 on non MoE. (https://www.reddit.com/r/LocalLLaMA/comments/1n79udw/comment...)

And while it has unified memory the memory is quite slow. 250GB/s compared to 500+ for M4 Max or 1800 GB/s for a 5090. So it's fast for a CPU, but pretty slow for a GPU.

(That said, there are not a lot of cheap options for running large models locally. They all have significant compromises.)

Using chip and pin is what makes the system secure. The PIN is verified by the chip on the card so it can't be skimmed. (You can copy the PIN of course, but you also need the same physical chip to clone a transaction.)

Signing the recipe is a useless security theatre. That's why it's not done in Europe.

Git won't catch new files the agent is adding. To get around that you can of course always add all new files, but then you'll potentially have your repo polluted with a bunch of temporary scratch files instead.

You can typically go back and edit git history. But it will require force push and breaking changes. And a few sacrifices to ensure that it doesn't make a mistake because then your repo is potentially broken.

Best way to do that is probably to have it work on branches and then squash merge those.

The problem is Dunning Kruger effect.

The people who shouldn't disable these security features tend to be the first to do so. And then complain the loudest when the enter the "find out" phase.

As someone who used to work for a bank building applications I would say no. This is definitely a feature companies and organizations like banks would request if it wasn't available.

There are a lot of scams targeting vulnerable people and these days attacking the phone is a very "easy" way of doing this.

Now perhaps there is a more forgiving way of implementing it though. So your phone can switch between trusted and "open" mode. But realistically I don't think the demand is big enough for that to actually matter.

As someone who developed in it at the time I found the reason it died was because they made new, slightly incompatible, versions every new Windows release.

After a while people got tired of doing updates.

From what I've seen immediate mode seems suitable for less fancy UI requirements. If you want to start having a framework solve things like animations and such then you'll probably end up with some form of retained mode.

Over my years making UIs I've found most of the bugs you get is due to incorrect state handling in applications. Having a framework you use which is opinionated and helps you solve that is pretty nice. (If your UI requirements are such that you need it.)

After being stuck in XML land for making UI for almost two decades I couldn't be happier to leave it behind.

Spreading your code over multiple languages means you now have a bunch new fun edge cases to deal with when things get stuck and don't update correctly. With everything in one language that means all debuggers and other tools work the same regardless of where in the app I'm debugging.

I've found it fascinating to see UI building come full circle. I started in code (C), then using various XML schemes and now back in code (Kotlin, Flutter, Swift). But a big part of why it's nice now is because C is a pretty horrid language for making modern applications.

On Android (and I'm pretty sure iOS) you have WYSIWYG editors for UI layouts for jetpack compose and flutter (and on iOS it would be Swift UI).

These all go direct to code instead of XML or some other extra layer of code.

What I would probably focus on is better integration with Figma and similar tools. Use that to do the WYSIWYG part and then generate corresponding code (possibly with LLMs).

The biggest limitation you tend to have in WYSWIG is that at some point you really need to have the true data the user sees to ensure everything looks good. That becomes a bigger hassle than coding the UI manually. (Particularly when doing multi-platform things.)

Kotlin multiplatform has been around for some time if you want that. But I guess it makes sense to be able to approach it from the other end as well if you're mainly an iOS shop.

It's "Tools as a service", so it's really trying to make tool calling easier to use.

I suspect that Waymo car's could operate in a lot more areas than they do. The issue is that Waymo are trying to sell the service of safe travel and not a car with an addon you can pay for which doesn't actually work.

In other words, since they accept liability for their cars it's not in their interest to roll out the service too fast. It makes more sense to do it slow and steady.

It's not really a strong argument that their technology is incapable of working in general areas.

My experience with this is that it is vital to have a system where the system can iterate on its own.

Ideally by having a test or endpoint you can call to actually run the code you want to build.

Then you ask the system to implement the function and run the test. If it hallucinates anything it will find that and fix it.

IME OpenAI is below Claude and Gemini for code.

The people who are used to making models are probably not the same as are building levels.

At least when I last tried Blender the UI was a lot harder to use than something like Unreal or Unity editor.

I'd probably look at the Godot editor first.

But I do like the idea of trying to find an existing editor in the same domain and modifying that rather than building from scratch.

For me personally there was a very big step going from copilot to cursor. Much bigger than going from "normal" programming to copilot.

Copilot seems to perpetually be 3+ months after the competition.

When I've been experimenting with Cursor I've found that one of the essential parts is the agent MUST be able to close the loop itself for it to be useful in any real sense.

So if you are having the agent write a script to call an endpoint and do something with the result. Have it write a script that does that and then ask it to run the script itself. It will then be able to close the loop by itself and iterate on the code until it does what you want (usually).

It is also very useful to ask the agent to write a test for something and then run that test to close the loop in the same way.

You could do that. But then you need to explain to the LLM how to do the work every time you want to use that tool.

And you also run into the risk that the LLM will randomly fail to use the tool "correctly" every time you want to invoke it. (Either because you forgot to add some information or because the API is a bit non-standard.)

All of this extra explaining and duplication is also going to waste tokens in the context and cost you extra money and time since you need to start over every time.

MCP just wraps all of this into a bundle to make it more efficient for the LLM to use. (It also makes it easier to share these tools with other people.)

Or if you prefer it. Consider that the first time you use a new API you can give these instructions to the LLM and have it use your API. Then you tell it "make me an MCP implementation of this" and then you can reuse it easily in the future.

LLMs work with tokens, not letters. So that's not going to work.

It might work in an agent system where it can make and execute code to solve problems.

I have a M4 Max 128 GB MacBook Pro. It's about the same speed as my 3090 when running in LM Studio.

These system work by taking a list of tokens (basically words) and the "model" and send that to a function which returns one new token.

You add that new token to the list of tokens.

Repeat with the new list of tokens.

That's how these systems work. We don't know exactly how the model works (wrt input tokens) but even that is a simplification. It's not magic. Just maths that's too complex to understand trivially.

If you want to get a better understanding of this I recommend playing around in the "chat playgrounds" on some of the engines.

The Google one allows for some free use before you have to pay for tokens. (Usually you can buy $5 worth of tokens as a minimum and that will give you more than you can use up with manual requests.)

https://aistudio.google.com/prompts/new_chat

This UI allows you to alter the system prompt (which is usually hidden from the user on eg ChatGPT) and change to different models and change parameters. And then you give it the chat input similar as any other site.

You can also install a program like "LM Studio" and that will allow you to download models (through the UI) and run locally on your own machine. This gives you a similar interface to what you see in the Google AI Studio but you run it locally. And with downloaded models. (The model you download is the actual LLM which is basically very large amount of parameters you combine with the input tokens to get the next token the system outputs.)

For a more fundamental introduction to what all these systems do there are a number of Computerphile videos which are quite informative. Unfortunately I can't find a good playlist of them all but here's one of the early ones. (Robert Miles is in many of them.) https://www.youtube.com/watch?v=rURRYI66E54

I've had cursor suggest changes for debugging (like prints), you do have to copy paste the results though.

I've also had cursor generate a CURL request to test an endpoint and determine what responses it had so it could generate a struct to store the result.

So I'd say they are getting there.

This is s side comment, but banks don't actually pay out the ass to get people to work on COBOL. That's a myth. They actually just pay people (sometimes programmers, sometimes not) to learn COBOL. (And pay them fairly unconpetetive salaries, because COBOL is not really all that useful.)

They do pay a lot of money to the people who program COBOL and maintain their archaic IBM mainframes. But that's mostly because it's basically a giant proprietary stack (hardware-OS-softare).

If you can't figure out something innovative and ground breaking to do as a project try to copy something which already exists and you find useful.

It can be an app or website, or something else. By trying to replicate something you will begin to learn and understand what the limitations and opportunities in these tools are.

Developers hanging out on eg Hacker News are the very tip of the wave. 99% of all developers don't visit sites like this. It will take a long time (10+ years.) before AI moves through all fields. Companies which are inherently software focused will be first. But that's not where the long tail of software is.

The long tail is in companies which does no software currently. Places where software is something you buy to do inventory or keep track of invoices or time sheets.

I use these tools everyday now and they are both magically awesome and stupendously stupid, sometimes in the same reply. Things always change less in the short term (say less than 5 years), and more in long term (10+ years). Just as it was with smartphones, arguably the most recent big revolution which is now integrated everywhere.

To reiterate. To find a project chose something which already exist and you use and have use of. Then copy it. Then improve it (to better for your use.)

Personally I think this new revolution will make software exponentially more abundant and infinitely customizable and adaptable to individuals. My guess is also that in 10-20 years we will have more people than today who do "programming".

A lot of tasks which are now hacked together in Excel sheets will be a lot easier to make into "proper" programs.

In this world the people who know enough about many things to efficiently use agents to accelerate them will be the most valuable.

There have been stores which accepted bitcoin as payment. Webhallen.com is an electronics store in Sweden which at least used to accept bitcoin (not sure if they still do).

As a general rule it's not very convenient to do so though since the value can fluctuate. (Which naturally all currencies do, but it would be kind of like paying with USD in the EU. You could do that, but most stores are not interested in the extra hassle of keeping track of multiple currencies.)

It is also not uncommon for services like VPN or IPTV streaming ("pirate streaming") providers to accept crypto.

These are language models, they are not designed for producing image output at all. In a way it's impressive it can even produce working SVG code as output. Even more sonthst it vaguely resembles a bird on a bike.

I have never been in a Lidl Store I would consider "first class". It's not bad products but they have a few factors that severely limit the impressions.

They mostly only carry their own brand with extremely limited options. If you want orange marmalade they have one style from one brand (their own). If you want some other style, find a different store. (This is the case with most of their items besides a small additional selection of eg certain candy, but still very sparse compared to other stores.)

Most items are set up in bulk on the cargo carriers it comes on. Although this is probably also a lot for show to give a "price consious" appeal.

They tend to have a significant part of the store dedicated to whatever random garbage they got a good deal on. Sometimes it's a boardgame, sometimes it's a stuffed animal, sometimes it's a power tool. Usually decent quality, but rarely the best.

For me I put Lidl in the class of store where you have to make multiple trips to different stores. If I'm cooking I can rarely go only to Lidl. You usually go there to buy some of the stuff you need but then get the rest at some other store with more items.

Well, I think most games err to the "quick and dirty" side of things. Tracking down intermittent issues in code like this must be horrible, if not impossible. Leading to delayed projects, buggy releases and developer burn out.

There's also one on Software Engineering Daily, quite interesting and it made me consider trying it out just to see what the building tools are like.

https://softwareengineeringdaily.com/2019/12/18/roblox-engin...

Ehhhhh...

I've use VisualStudio every day for the past years. And honestly there are a lot ideas they could borrow from other tools like Eclipse.

Some examples that I personally feel are ways that C# in VS is lackluster compared to Java in Eclipse.

No proper automatic build. While the stuff in Eclipse is a bit too aggressive sometimes it's really one of those really cool features to have.

Errors, warnings, other tools like StyleCop don't show errors inline with the code. I mean, really? I have to scroll through the error list and click?

It also doesn't really seem to understand the code it's processing. I'd expect that when I'm working on Silverlight with C# that the premiere tool from Microsoft should be able to handle things like refactoring properties in C# where it would update automatically in XAML.

And this is not even mentioning the complete trainwreck that is MSDN. Perhaps you can understand it once you really get into it, but I find it extremely confusing the way it mixes different frameworks with similar but subtly different functionality. (Eg a lot of things in Windows Phone are similar, but not identical, to features in .Net and trying to find the "chain of classes" that are relevant to your code is not nearly easy enough.)

Some of these things are probably things you can fix with extensions like Reshaper. But in the end you'll pay hundreds of dollars to compete with something that is free in Eclipse.