rutierut

In this thread I’m seeing a couple of the typical skill arguments. Skills don’t pollute the context window and, skills are more flexible and composable.

In this specific post the SQL example gets around this by being super low-level, inherently composable, and low context consumption itself. I’m not sure this works for everyone.

Looking at the Linear MCP for example, they basically just want to expose an API surface and provide 32 MCP tools as a result of that. My agents do run into composability problems here all the time and fall back to the tool they included which allows you to run arbitrary GraphQL. This takes a couple tries usually. I don’t really know what the sensible alternative to this is.

Cloudflare switched to “code mode” which basically provides 2 MCP tools: search and execute. Both take a TS arrow function, the first allows you to programmatically search a TS api spec, the second to compose and run the methods in that spec.

I think this is a very interesting approach, and certainly better than providing the ~1000 actions as MCP tools but I feel the jury is still out on exactly how well this pans out.

Any other tips, I have basically the same thing.

As someone who has used Deno on multiple hobby projects, I’m convinced Deno is where the JS ecosystem should be heading.

Professionally though, it’s complicated recommending it outside of specific and mostly tightly scoped use-cases. At some point the project just changes direction because of business reasons and you need node.

Deducta | Senior Frontend Product Engineer | London, UK | Hybrid (3 days/week in office) | £150k+ cash/equity

We’re a data intelligence platform for global enterprises. We are well funded, growing fast, and solving the most pressing data challenges for some of the largest and most recognizable companies in the world. The capabilities of SOTA LLMs enable us to this at an unprecedented scale and success rate. We’re looking for a Senior Product Engineer (Frontend) to shape what our product looks like and how it’s built. Specifically:

- Lead on the deployment of LLMs as an enhancement/substitute to traditional interfaces

- Own feature developments and fixes across the frontend from ideation to deployment

Stack: Vue.js, Tailwind, Tanstack

Contact: kimathi [at] deducta dot ai / Apply: https://apply.workable.com/deducta/j/8ADBB3795A/

This might have changed but Warp was not able to do this without “warpifying” the SSH host.

CLI-first inventory management for EE parts. https://willwillems.github.io/tray

Fully local, hobbyist friendly, agentic workflows work great with it since it’s just a CLI.

Deducta | Senior Frontend Product Engineer | London, UK | Hybrid (3 days/week in office) | £150k+ cash/equity

We’re a data intelligence platform for global enterprises. We are well funded, growing fast, and solving the most pressing data challenges for some of the largest and most recognizable companies in the world. The capabilities of SOTA LLMs enable us to this at an unprecedented scale and success rate.

We’re looking for a Senior Product Engineer (Frontend) with strong design skills to shape what our product looks like and how it’s built. Specifically:

- Lead on the deployment of LLMs as an enhancement/substitute to traditional interfaces

- Own feature developments and fixes across the frontend from ideation to deployment

Stack: Vue.js, Tailwind, Tanstack

Contact: kimathi [at] deducta dot ai / Apply: https://apply.workable.com/deducta/j/8ADBB3795A/

My workday is fully automated now using this tool that I’ll sell you…

I know what you mean, biggest gripe I have with this game is how important it is to play the meta and boom early if you want to win at anything other than easy mode.

To be fair, for major features 30m to an hour isn’t out of this world. Browser testing is critical at this point but it _really_ slows down the AI in the last 15% of the process.

I can see overnight for a prototype of a completely new project with a detailed SPEC.md and a project requirements file that it eats up as it goes.

Poster is not from the US and has no interest in relocating due to the current political situation. This is a common sentiment among non-US but western professionals right now. Had the location been different, they probably would have been interested.

I think it’s very typical of post-rat and thus Aella that these things can exist at the same time and don’t invalidate or negate each other.

I’ve been using Warp (for the AI features) for a while now, but less and less these days. They’re way too agile with the UI/UX, things change around too much for it to be what it is supposed to be.

The first part of your post sounds almost like an ad for the Bambu Lab P1S. The second part sounds more like the Prusa CORE One kit (build volume is not a perfect match).

I really wouldn't bother buying anything else as a beginner. Pick between these two.

It's a weird thing in 3D printing right now that if you don't have the open source stuff as a requirement you get better print quality and reliability for half the price with Bambu Lab.

I really like Trigger but I would have loved some way to trigger it from the FE, it was a small part of our use case but it sucked having to implement a proxy for it.

I see a couple of people here bashing on the practicality of this project. That's obviously not the point, it's an interesting weird use case that's more explorative/educational than practical. I thought it was an interesting and inspiring read!

We're protecting the global poor by not letting them work for us is such a backwards piece of rhetoric that I'm always surprised by how casually it's brought up in stories like this.

DDG was such an absolutely subpar experience for me (used it for 6-8 months) that it wouldn't surprise me if nobody would _actually_ say it's better in a blind test. They just don't like Google.

Kagi on the other hand really blew me away, I have no idea how they're doing it, but I'm only using Google for one thing these days (besides G Shopping & G Maps) which is when I want to use ad spend as proxy for quality/reliability of the company (for example when I'm looking for a local moving company in my city).

> Additionally, a lot of water is also used in cooling for the servers that run all that software. Per conversation of about 20 to 50 queries, half a litre of water evaporates – a small bottle, in other words.

What? I have no idea about server farm cooling, can anyone explain this to me?

Indeed, an article making the opposite point would probably make more sense.

This article even opens with the point that smaller languages have less Wikipedia content and somehow blames "Tech". A language having less articles is a people/popularity problem. You being able to read them anyway through machine generated translations is a tech thing.

Yes that whole section is extremely off:

> How do we know that $7 trillion invested into LLMs and their infrastructure would not simply exacerbate those costs, grinding down content creators, women, and the environment, undermining democracy, destroying jobs, etc?

“Destroying jobs” is pretty much a _good_ thing. I think we’re all pretty happy we don’t need dedicated people anymore for a lot of tedious tasks that we take for granted these days (washing clothes etc)

“Widespread impact of AI” I know it’s helping people a lot already but is it actually causing layoffs?

> Overall, this report just seems like fodder for the anti-TikTok crowd,

Can confirm, I do belong to that crowd and I did share it after reading it. An hour later, it dawned on me that this report is maybe kinda cool, but really isn't enough to draw any conclusions from and I should not have shared it.

The uncompressed WEBP image looks terrible to me with a lot of banding on Safari mobile. Did the author accidentally switch images or is Safari doing some “optimization”?

Probably if you’re working a lot with photography compression artifacts start to become a real eyesore. Especially the first lower quality webp image does look like shit to me but I also realize a lot of other people would not consciously notice.

The banding is just not supposed to be there.

I cannot think of a _single_ low-margin global corporation that can get away with it that hasn’t made the trade-off where they accept some of their customers will be treated horribly in order to save _a lot_ of money.

I’ve currently got a DHL package in limbo that’s worth about the same and the main function of their customer service department seems to be to indicate that trying to consume customer service resources is not going to help you. Employees will give you completely contradictory and verifiably incorrect information.

I appreciate the sentiment but the idea that people support policy that’s good for them isn’t much better either.

For an obvious example check how many people support the Affordable Care Act vs Obamacare.

> I was happily avoiding full time employment. I took this job because I believe that OpenAI is one of the most important companies currently in existence. When the board shared the situation and asked me to take the role, I did not make the decision lightly. Ultimately I felt that I had a duty to help if I could.

An external CEO that's stepping in not because they want to but because they feel they "need" to doesn't sound like a recipe for success.

It's a wonder that obnoxious hit pieces like this still get made in this day and age. The first paragraph is filled with disingenuous strawman rhetoric.

> colonize space, plunder the vast resources of the cosmos

The author obviously tries to draw a parallel between inter-earth colonization and plundering to make longtermism and by proxy AE look bad.

I'm not an EA but I've never met people more receptive to criticism as they are. This is a group of people, uniting around a desire to do good, actually going through with it, and somehow catching a huge amount of flak for it.

It was completely baffling to me how someone as brilliant as SA could conclude that "the risk of dying from the screening exam was 1/660".

However, the paper he references does contain the specific example:

> For example, a 40-year-old woman undergoing CT coronary angiography is estimated to have a 1 in 270 chance of developing cancer at a radiation dose of approximately 20 mSv

At this point I'm completely lost here.