burningion

I realize there are already a ton of comments, but I think you're missing the idea of _precision_.

Most people aren't precise when initially describing their problem.

Jumping straight into implementation skips the part where we refine and better define what it is we're trying to do, and think through the implications of those changes.

I suspect "trusting the model" doesn't really work at scale with finite resources.

Not making the argument of whether or not it's corruption, but the DOJ sought to provide a legal exemption for the XAI datacenter:

https://apnews.com/article/musk-xai-data-center-memphis-poll...

It's also cheaper to just pirate the author's work, which Anthropic has also done.

I wish this were higher up!

They pirated my work and now they want government protection from other people doing the same.

The splat of the studio has the perfect amount of detail. It looks like you're streaming from the camera direct to the computer usually? How do you check the progress / quality while capturing? Seeing the (great) results makes me more curious about the process of creating now.

The main point raised in the article is that these bots may void attorney client privileges.

But the real danger with these IMO is that they're turning casual conversations into a permanent record, and one that will be completely discoverable in court, should the company get into trouble later.

Ran this through Qwen3.5-397B-A17B, and the difference between 4 characters and 5 is wild to see:

> are the following parenthesis balanced? ((())))

> No, the parentheses are not balanced.

> Here is the breakdown:

    Opening parentheses (: 3
    Closing parentheses ): 4

... following up with:

> what about these? ((((())))

> Yes, the parentheses are balanced.

> Here is the breakdown:

     Opening parentheses (: 5
     Closing parentheses ): 5

... and uses ~5,000 tokens to get the wrong answer.

Thank you for posting this!

I suspected my work was in the dataset and it looks like it is! I reached out via the form.

This is why an on device browser is coming.

It'll let the AI platforms get around any other platform blocks by hijacking the consumer's browser.

And it makes total sense, but hopefully everyone else has done the game theory at least a step or two beyond that.

I agree, but I also didn’t create redis!

It’s a tough bar if LLMs have to be post antirez level intelligence :)

or... maybe there's something to people being skeptical of datacenters?

https://www.reuters.com/business/environment/musks-xai-opera...

https://www.politico.com/news/2025/05/06/elon-musk-xai-memph...

> In just 11 months since the company arrived in Memphis, xAI has become one of Shelby County’s largest emitters of smog-producing nitrogen oxides, according to calculations by environmental groups whose data has been reviewed by POLITICO’s E&E News. The plant is in an area whose air is already considered unhealthy due to smog.

Had this set the precedent of working with the community, and _not_ breaking the law, I think we'd be in a better place all around.

Similarly, Amazon tried to take the excess nuclear power, without paying back into the electrical grid infrastructure, and got denied in 2024:

https://www.utilitydive.com/news/ferc-interconnection-isa-ta...

and again in April of 2025:

https://www.ans.org/news/2025-04-16/article-6937/ferc-denies...

Yeah it's live on the pricing page:

https://www.anthropic.com/pricing#api

Opus 4 is $15 / m tokens in, $75 / MTok out Sonnet 4 is the same $3 / MTok in, $15 / MTok out

I know everyone likes to abstract away the costs associated with AI data centers.

But let's look at what has happened with Grok, for example:

From May 6, 2025

https://www.yahoo.com/news/elon-musk-xai-memphis-35-14321739...

>The company has no Clean Air Act permits.

> In just 11 months since the company arrived in Memphis, xAI has become one of Shelby County's largest emitters of smog-producing nitrogen oxides, according to calculations by environmental groups whose data has been reviewed by POLITICO's E&E News. The plant is in an area whose air is already considered unhealthy due to smog.

> The turbines spew nitrogen oxides, also known as NOx, at an estimated rate of 1,200 to 2,000 tons a year — far more than the gas-fired power plant across the street or the oil refinery down the road.

The details are in the specifics here. People are _already_ feeling the effects of the AI race, the consequences just aren't evenly distributed.

And if we look at the "clean" nuclear deals to power these data centers:

https://www.reuters.com/business/energy/us-regulators-reject...

> The Talen agreement, however, would divert large amounts of power currently supplying the regional grid, which FERC said raised concerns about how that loss of supply would affect power bills and reliability. It was also unclear how transmission and distribution upgrades would be paid for.

The scale of environmental / social impacts comes down to how aggressive the AI race gets.

The linked url for this seems to be spam maybe? None of the urls work on the page? This page should probably be replaced with any of those links.

MCP was originated at Anthropic, and has been adopted by OpenAI, Github, Cloudflare, and more.

It's completely open, with active engagement and direction from the community:

https://github.com/modelcontextprotocol/modelcontextprotocol

I don't think it would have such wide adoption so rapidly if it were an "over-engineered implementation".

Yes, rerun does replay, that was my main use case when prototyping.

They've since raised more funding recently, and have larger use cases in mind for robotics: https://rerun.io/blog/physical-ai-data

I've spoken with members of the team, and they're all great. Wouldn't hesitate to use the product / work with them anywhere.

Tailscale is a great. I think of it as a swiss army knife for easier routing and connectivity.

I use it in projects to stream internet / connectivity from my phone to the NVIDIA Jetson line, making my robotics projects easily accessible / debuggable:

https://github.com/burningion/bicyclist-defense-jetson?tab=r...

So I think there's an assumption you've made here, that the models are currently "60-80% as good as human programmers".

If you look at code being generated by non-programmers (where you would expect to see these results!), you don't see output that is 60-80% of the output of domain experts (programmers) steering the models.

I think we're extremely imprecise when we communicate in natural language, and this is part of the discrepancy between belief systems.

Will an LLM model read a person's mind about what they want to build better than they can communicate?

That's already what recommender systems (like the TikTok algorithm) do.

But will LLMs be able to orchestrate and fill in the blanks of imprecision in our requests on their own, or will they need human steering?

I think that's where there's a gap in (basically) belief systems of the future.

If we truly get post human-level intelligence everywhere, there is no amount of "preparing" or "working with" the LLMs ahead of time that will save you from being rendered economically useless.

This is mostly a question about how long the moat of human judgement lasts. I think there's an opportunity to work together to make things better than before, using these LLMs as tools that work _with_ us.

When you work for most public corporations, you aren't allowed to bring personal devices linked to company servers to specific countries. You need to bring a burner device instead, because you are perceived as a target for corporate espionage.

This is like that, except the government and the type of people on the list are even better targets for their personal devices. The government has strict rules about secrecy and communication for military operations, and strong punishments for not following these protocols, because they can lead to a loss of life.

This is a different sort of "unsecure". The platform itself may be "secure", but the device, being in public where someone could take a picture of military secrets, etc. isn't.

That example code on DeepSeek doesn't actually include the logic to call a weather API? It just puts a fake answer back in, and you've got to handle the process manually.

The pyproject.toml in the Model Context Protocol example is just showing the new, "best" way to distribute and install Python projects and dependencies. If you haven't used uv before, it makes working with Python projects substantially better.

The Model Context Protocol server lets the model autonomously use the tool and incorporate its result. It's a much cleaner (imo obviously) separation of tool definition and execution.

I think another category of error that Simon skips over that breaks this argument entirely: the hallucination where the model forgets a feature.

Rather than the positive (code compiles), the negative (forgets about a core feature), can be extremely difficult to tell. Worse still, the feature can slightly drift, based upon code that's expected to be outside of the dialogue / context window.

I've had multiple times where the model completely forgot about features in my original piece of code, after it makes a modification. I didn't notice these missing / subtle changes until much later.

I've been using Mongo while developing some analysis / retrieval systems around video, and this is the correct answer. Aggregation pipelines allow me to do really powerful search around amorphous / changing data. Adding a way to automatically update / recalculate embeddings to your database makes even more sense.

I spoke with a chicken farmer last week.

The government comes in and takes over. You don’t get to decide, they kill all your chickens and cut you a check.

No, these are not only the laws they allegedly broke.

They created a project named Greyball to identify law enforcement and mislead them.

They created a kill switch for the event of a government raid to gather evidence.

They ordered and then canceled rides on competitor apps.

They tracked journalists and politicians...

The list goes on and on: https://en.wikipedia.org/wiki/Controversies_surrounding_Uber

This argument ignores the fact that there were other alternatives to Uber at the time, ones that didn't break the law! Believe it or not, there were multiple ride hailing apps on the iPhone, but none were as great at accumulating capital or breaking the law without recourse.

Yeah, agreed. The essay I linked to adds the cost of choice for a medium against a potential reward too.

I think of "social" media as a fundamentally isolating experience.

On modern social media platforms, each person gets their own, personalized feed.

As their feed becomes more personalized, it gets more and more isolated from a common, shared cultural context.

According to Wilbur Schramm's theories of communication, human to human messages must occur over a shared cultural context, and social media shrinks that shared cultural context.

LLMs as a platform could shrink this shared cultural context even more, unless we become more deliberate about fighting back against this "shared experience" shrinking:

https://www.makeartwithpython.com/blog/social-media-is-the-f...