leblancfg

That's also where this map would help, it can definitely degrade gracefully. If I know I have two right-hand turns coming, I can take the decision for myself.

I need to know what lane I'll need to be in next so I can plan in advance.

Great idea and demo. My main gripe when using Google Maps in the car is lack of information about the next turns — when do I change lanes, and to which one? This could be natively addressed here, especially if the blue navigation line was a little thinner and showed which lane(s) the driver needs to take.

Author here: funny you should mention that! Same idea has been on my mind for years at this point.

To me, the holy grail of sports training is to have simple but powerful enough models to estimate power data from dozens of movements, and fit various power curves that match real-world training.

So much promise there.

---

As for Stryd: That's so interesting. I'd imagine a lot of the same parallels to cycling to apply? Once you have your running CP/FTP, I'd imagine you could follow the same cycling training plans (relative to your FTP) for the same intended stimulus.

I'm curious though, can you see the Watts/zone/W'bal/etc. values on a watch or something when you're running?

Thinking back, I should also add: if you're interested in how hard a single hit is, then yes joules!

I'm approaching it as a workout, where you're doing dozens of hit. Good idea to have an individual strike joule view to the display though! "carnival mode"

I didn't look into or find any myself, but "power meter enabled" sports like cycling, rowing, etc. have a way to measure the strain/fatigue of efforts relative to the athlete's fitness (FTP). Same would apply here.

There are various methods to measure that strain, e.g. Strava, intervals.icu, TrainerPeak all have their slightly tweaked implementations.

100% agree. And the issue with med-ball slams is they also can just be dropped in a testing setting, and the only power needed to be generated is on the "up" cycle. I wouldn't even trust my own med-ball slam benchmarks from one month to the next, as I'm not confident I slammed them with the same intensity.

I'm curious though, wouldn't a Concept2 ski-erg be the best machine for that job?

>track your one rep max force [...] especially for sprinters

Yeaaah that's the ticket =) One level of abstraction higher: you can think of that 1RM as the leftmost point on the max power curve, i.e. max force you've historically generated, from 5secs all the way to e.g. 60 mins [1].

(1) https://www.trainingpeaks.com/blog/4-key-uses-for-the-power-...

Both are needed, but for an athlete watts is by far the more useful feedback for training, and following the intended stimulus of workouts.

As a workout, "bike 30 km" is vague and meaningless. "bike 30km in 1h" is better, but not specific to the athlete: some have better cardio than others. Most serious cardio workouts are given in zones: "bike 1h at 60-65% of your functional threshold power" [1] – this particular workout would also be known as "zone 2 training session".

(1) https://www.trainerroad.com/blog/what-ftp-really-means-to-cy...

Aye, author here. Pleased it's getting visibility ofc, but ABSOLUTELY not expecting to get traction this early.

Please pardon the AI-generated placeholder images and some of the text at https://intensity.systems, I'm still very actively working on that.

2m is too small

uv's pip interface is like dipping one toe in the bathtub. Take a minute and try on the full managed interface instead: https://docs.astral.sh/uv/concepts/projects/dependencies. Your commands then become:

- uv add <package_name>

- uv sync

- uv run <command>

Feels very ergonomic, I don't need to think much, and it's so much faster.

My Samsung computer monitor is also the stuff of nightmares. Same story: useless "smart" UI features. I'm told I can use it as a dozen different things. But it sucks as a computer monitor.

Not cheap either!

Fine tuning was never really hard to do locally if you had the hardware. What I’d like to read in an article like this is more details into why they’re making a comeback.

Curious to hear others’ thoughts on this

> When did people start hating node

You're kidding, right?

Same comment as above. From your article:

> there are some underlying factors skewing these results, such as: > * grammar schools are more likely to be single-sex > * co-educational schools have a higher proportion of poorer pupils > * girls are more likely to get good results

I'm not familiar with the Australian education system or this study's design, but at first glance, this quote

>The report, commissioned by the Alliance of Girls’ Schools Australasia, was conducted by Macquarie Marketing Group using OECD data

reads more to me like "we found that all-girl private schools are better than the average of public and private schools", and the obvious reason why is probably *because they're private schools*, and not because they're all-girl.

The amount and extent of data that is available out there by brokers for purchase by literally any company is *mind-boggling*. However bad you think it is, multiply that by 10.

I hate the idea of MCP servers being their own OAuth providers and IMO I wouldn't use this. MCP servers just shouldn't have to handle OAuth user management. The only reason PostgreSQL is needed here is to store OAuth tokens for its users. It works though ¯\_(ツ)_/¯

The draft spec[1] – soon to be published AFAICT – flips this on its head and says "no, the MCP server is an OAuth2 protected resource, just go and talk to this authorization server and I'll accept its tokens". The MCP server doesn't need to handle anything related to auth. That's much saner, and would recommend people build towards _that_ particular future in mind.

But also broadly speaking: I strongly believe the mental model needs to move to MCP servers are URLs. It should just be extra endpoints and controllers on top of your app. I don't want to `npm install mcp-your-app` or go through someone else's implementation, I want to hit `https://your.app.com/mcp` directly.

I guess it's fine having the separate infra for an MCP server used to connect to 3p services... but fingers crossed, your 3p services starts exposing MCP endpoints directly.

edit: Add URL, fix words

1. https://modelcontextprotocol.io/specification/draft/basic/au...

According to that link that's for dispersible form (gas, powder) – it's up to 7kg in solid form ¯\_(ツ)_/¯

Love his series, but you really have to keep in mind that his testing methodology is "reductionist": a microphone's quality can't just be distilled to its EQ response. Phase, transient response, clipping, etc. are also part of the equation.

edit: My "case in point" moment is Jim's (excellent!) DIY amp section in this video: https://youtu.be/wcBEOcPtlYk?si=jkehIfyo6AgeTLUo&t=918. Its EQ indeed sounds like the big names, but I'm sure you'll also notice how its dynamics also sound so thin. That's likely because solid state clipping != (saggy) tube overdrive.

That’s great, but would love to hear samples to see if it’s any good!

In this analogy, the AI hardware is not the public good – the AI models are.

After an innovation phase there is an implementation phase. Depending on the usefulness of the innovation, the integration with existing systems takes time. It is calculated in years, tens of years. Think back on the 80-90s, where it took years to integrate PCs into offices and workspaces.

From your comment, is sounds like you think that the implementation phase of LLMs is already over? And if so, how do you come to this conclusion?

Here's a thought experiment. Think back on how that statement would have sounded like to past-you, 3 years ago. You would probably have dismissed it as bullshit, right? We've gone a long way since then. Both in terms of better, faster and cheaper models, but also how they're being intertwined with developer tooling.

Now imagine 3 years from now.

OK that’s neat but boils down to “here’s a cool game”. I was expecting a double blind study that measures some concentration metric vs a placebo intervention.

Please do whatever you feel like doing. If you like the aesthetic and it fits the overall design, why not? Don’t let some HN commenter tell you what you can’t do.

"opinionated" != "jerk"

Recently switched to WezTerm and I'm very happy. Was using kitty before that – loved the set up and simplicity coming from iTerm2. WezTerm is leaps and bounds better in terms of what comes out-of-the-box. My terminal config is short enough to sit all in one screen on my editor. After that, the terminal just... gets out of the way and I don't need to think about it.

But the straw that broke my back with using kitty was, I'd end up encountering issues or trying to recreate some of iTerm2's features, only to end up time and again on kitty's maintainer's terse and dismissive comments.

e.g. IIRC his answer to "How do I set up tmux with kitty?" was something like "Don't, tmux is dumb" and closing it. Eventually I gave up.

I read the following:

> The focus of this study was not to analyse the prevalence of the occurrence of the median artery in relation to ethnicity, geographic origin or variations by sex, but to identify the global trends in its occurrence.

and thought "well that explains the astronomical precision of the p-value, they didn't take ethnicity into account!". You would expect to see variation prevalence vary by ethnicity, no big news here.

But no! Thinking about it more, that's exactly what is says on the tin: global ethnic populations changed in the last ~140 years, and with it, the prevalence of generic variations. Makes perfect sense ¯\_(ツ)_/¯

edit: I mean... assuming the causal link ofc. I'm assuming this makes more sense than some kind of evolutionary pressure that is selecting for forearm median arteries.

For me kitty’s main upside is its configuration dotfile that can be version controlled. I used iTerm2 for years, but moving around multiple computers makes it harder.