dimfeld

Claude Code works similarly to Aider in that they both run in the terminal and write code using LLMs, but it shares no code with Aider as far as I know. Aider is written in Python and Claude Code in Javascript, among other reasons to think that it is not derived from Aider.

The tools also work very differently when you're actually using them, with Claude Code doing more of an agent loop and having very different methods of putting together the context to pass to the model.

Google for AHA stack. (Astro, HTMX, Alpine) There was a great site by Flavio Copes that went into a lot of detail on using them together but it looks like it’s gone.

I agree the title was a bit confusing. There were actually two separate AI pendant products, both named Friend.

My understanding is that this is one of them, and they changed the name shortly after the other was released.

Edit: ah seems the title has been updated

LLMS tend to be pretty bad at answering questions about which one it is, what version, etc. You can put stuff into the system prompt to try to help it answer better, but otherwise the LLM has little to no intrinsic knowledge about itself and whatever happens to be in the training data shows up instead (which now is a bunch of ChatGPT output all over the internet).

I have no inside info, but it sounds like the key was inadvertently bundled into the client-side code. This could happen when using web frameworks that do both client-side and server-side rendering, if one of your client-side files imports something from a file that is supposed to be server-only, and contains the API key environment variable.

Some frameworks automatically detect this and fail to build if you do it, but apparently not all of them.

Only the weights themselves. There have been other models since then built on the same Llama architecture, but trained from scratch so they're safe for commercial user. The GGML code and related projects (llama.cpp and so on) also support some other model types now such as Mosaic's MPT series.

Corresponding statement from bit.io here: https://blog.bit.io/whats-next-for-bit-io-joining-databricks...

Thanks, I hadn’t seen that!

Hmm, I thought this only really applied when you use functions which Postgres can’t reason about. Hence the LEAKPROOF keyword to help with this case.

Either way, I think it underscores that RLS shouldn’t be used for business logic. I’ll update the page to emphasize that more.

Author here. This page was actually just some notes I threw together in a couple of hours combining some research on RLS with my own existing knowledge, so it's quite possible there are some inaccuracies.

My "source" for thousands of users not scaling well was simply a few comments I found scattered around; there wasn't really anything empirical to back up that line. I'll remove it. Thanks for the note!

Hey all, this is the first in series I'll be writing on working with geographic data. This one is targeted toward newcomers to the geographic data world, to set the stage for more advanced topics.

Looks like there are some demo videos on the author's Youtube channel: https://www.youtube.com/watch?v=NlRxRtGpHHk

I've seen some of the GitHub issues that Rich is referring to in those tweets. They're almost always cases where someone is asking for some feature or change in Svelte and the core contributors disagree that it fits with the design and vision of the project.

Then the OP tags Rich in the hopes that he will swoop in and and override the decision. It's almost always more of a tantrum than a real request for help.

As far as project progress, most of the recent work has been on SvelteKit, but I've been using Svelte for about a year now and it has steadily gained features during that time. It's definitely not in bugfix-only mode.

I wrote some about my experiences using Leaflet with Svelte: https://imfeld.dev/writing/leaflet_with_svelte

I found you have to do some extra work to keep track of things to convert from declarative style to Leaflet’s imperative style, but overall I actually had a pretty good experience with it. Feel free to ping me on Twitter (same username) if you have any questions.

I made FundTheRebuild.com yesterday as a small way to help people and small businesses who are suffering in the wake of the riots going on throughout the US right now. It's designed to bring attention to the GoFundMe campaigns that don't go viral.

All the sorting options focus on campaigns with the least number of donations, most money remaining toward the goal, and so on. The list is still fairly small, so if you know of any other GoFundMe campaigns that should be on there, you can submit them at https://fundtherebuild.com/add.

I had the same thing happen recently, and came across this Youtube video detailing a fix: https://www.youtube.com/watch?v=yFaLyoSQQo0.

Essentially, you open up the mouse and loosen the internal screws on the thumb button a little bit, which moves the board away from the rubber over the button and relieves whatever pressure is keeping it pressed. You need a screwdriver with a T5 Torx bit to open up some of the screws, but overall the procedure is pretty easy and my mouse has been working great ever since.

It's notable that the creator of Tailwind, Adam Wathan, is working on something similar [1]. Whether it fits the paradigm or not, there definitely seems to be a demand for such a thing.

On a recent episode of the Full Stack Radio podcast [2], Adam talked about his own upcoming Tailwind UI release and said that he hoped the utility class design of Tailwind would enable people to use the predesigned components as a starting point and easily customize them, should they desire.

[1] https://www.tailwindui.com/ [2]: http://www.fullstackradio.com/123

A lot of that is done in a simulator on the computer, which usually builds very quick once you get the environment and simulation scripts written.

The full build to a bit file that can be loaded into hardware is an NP-hard optimization problem, so as your design approaches the limits of either the desired clock speed or the amount of space in the chip, it can become very slow.

Perhaps we have different interpretations of those quotes, but I think they reinforce my point, which is that they take a “rules are rules” approach rather than thinking of if the rules actually make sense.

The article says “The World Bridge Federation (WBF) is recognised by the International Olympic Committee and as such abides by World Anti-Doping Agency rules.” It’s unfortunate that nobody quoted in the article seems to realize the absurdity of this action, but maybe staying in the good graces of the IOC is too important to them.

One example I’ve found is the trading software for Interactive Brokers. It’s a Java application with mostly custom (and fairly clunky) UI, and it has other Windows-isms like adding an icon to the desktop when you install it.

The libpostal developers have released bindings for a number of different languages, which can be found at the Github organization page: https://github.com/openvenues

GROUP BY ROLLUP support actually just arrived in Bigquery a week or two ago. Still a lot of things missing, but they are steadily improving.

I suspect that "But at that time (early May), I was interested in learning Rust" was a big consideration there. But I can also vouch for Go's excellent cross-compilation support.

Somebody opened a huge options position on EFX a couple weeks ago. No way to tell who, of course, but it certainly looks suspicious. I posted the chart here: https://twitter.com/dimfeld/status/905952247682547716

This is the approach I ended up using too. Works great.

Cayley's graph data layout is most similar to a Hexastore-style [1] triple store, though IIRC it doesn't do the full six-way index that the original Hexastore paper describes. The Redis page on secondary indexing [2] has a great quick intro to what this actually entails (search the page for Hexastore).

As you might guess from the Redis link, this style of graph lends itself well to KV stores, so the answer to your question #1 might be that it's a long-term decision, but the style of graph is really designed for a KV store anyway. But I haven't discussed this at all with the Cayley devs so I can't actually speak for them.

I'm using it with the BoltDB backend and have been pleased with the performance overall. I haven't looked at the backends for more complex databases like Postgres in detail, but it does appear that the backend interface has potential for predicate pushdown as well. The repository's graph directory [3] contains the various backends if you want to check it out. Overall it doesn't look very difficult to add another backend type, but I haven't tried it yet. Looking at the existing SQL backend, it appears to already support MySQL, PostgreSQL, and CockroachDB (but I've tried none of these with Cayley).

[1] http://www.vldb.org/pvldb/1/1453965.pdf [2] https://redis.io/topics/indexes [3] https://github.com/cayleygraph/cayley/tree/master/graph

With a workload of text editing, compiling code, and light web browsing I never saw the new MBP switch to the discrete GPU when I looked at the activity monitor, and I was still getting battery life or 4-5 hours at best (much improved over the past few days though, for some reason). So I suspect this particular issue is not related to discrete GPU switching.

This fits with what I've seen. Got the new 15" a few weeks ago, and transferred my old system over to it. Battery life was 5 hours at best, usually less (actual usage, not estimated). But then just a day or two ago I'm suddenly getting 8-9 hours with no problem, no change in usage or applications. I don't have much data in iCloud either, as far as I know.

If it is related to background indexing, I'm surprised that there doesn't seem to be much visibility into it. None of the Activity Monitor tabs or tools such as htop revealed any particular heavy CPU/disk usage to which I could attribute the poor battery life.

Hmm, interesting. For what it's worth, I haven't had that problem.