I've had Tesla FSD come up in three independent conversations in the last week, as in "whoa isn't it crazy how all these engineers are riding around in FSD while paying zero attention to the road?". It seems like this is becoming common place.If you do this, shame on you. You are endangering others. FSD is not safe for autonomous operation.
My take on this is vamp: https://dlants.me/vamp.html
I've been using it with joy on all of my side projects, and feel quite miserable when I return to react.
Would you say this is accurate? Have you had challenges generalizing to other sorts of work (you mentioned you're spreading it out to other repos)?
- the relatively crude tactile and proprioceptive sensing apparatuses of robots when compared to humans
- the limited availability of multisensory, perception-action coupled training data
Genuinely curious!
Blooms 2 sigma, like many studies that are thrown about, is not as much of a slam dunk as people claim that it is. Have you read the paper or any efforts to replicate?
And wrote about my thoughts on the relevance of nvim here https://dlants.me/ai-whiplash.html
It's been many months, and I thoroughly prefer my harness inside of nvim as my day to day development environment. Using Claude code or cursor makes me feel very removed from the code.
Exploring code, gathering context and tweaking prompts/giving guidance to the agent are very much enhanced by neovim.
The biggest boon has been the fact that agents make customizing neovim a lot easier. Writing new bindings, config, and even building novel plugins.
Here's a few that I built that fit into my workflow:
- a tool for reviewing commits / branches / wip. https://github.com/dlants/glean
- an iterative grepper https://github.com/dlants/shuck
- a file picker that renders in your current window (like oil) and uses more intelligent signals for ranking (like frecency) https://github.com/dlants/needle
Recently added support for scripts (like Claude code workflows) and been iterating on the UI for that a bunch.
I also ended up wanting other customized tooling - a more streamlined way to grep, find files and review code that my agent has written. So I wrote a few plugins for that : needle (finder with UI and sorting functions that suit me better), shuck (interactive grepper that has a workflow around refining grep commands) and glean (a review tool that lets you mark parts of the code as seen, leave comments, view diffs commit by commit or collapsed, etc). https://github.com/dlants/dotfiles/tree/main/nvim/lua
These are all in various states of experimental and mostly just for me, but a few of my coworkers and friends have been using magenta and like it.
I guess I kinda see it... it makes sense from both points of view (xAI needs data + places to run their models, cursor needs to not be reliant on Anthropic/OpenAI).
I think I don't see it working out... I just don't see an Elon company sustaining a culture that leads to a high-quality AI lab, even with the data + compute.
Fun observation - almost every coding harness (claude code, cursor, codex) uses a find/replace tool as the primary way of interacting with code. This requires the agent to fully type out the code it's trying to edit, including several lines of context around the edit. This is really inefficient, token wise! Why does it work this way? Because the LLMs are really bad at counting lines, or using other ways of describing a unique location in the file.
I've experimented with providing a more robust dsl for text manipulation https://github.com/dlants/magenta.nvim/blob/main/node/tools/... , and I do think it's an improvement over just straight search/replace, but the agents do tend to struggle a lot - editing the wrong line, messing up the selection state, etc... which is probably why the major players haven't adopted something like this yet.
So I feel pretty confident in my assessment of where these models are at!
And also, I fully believe it's big. It's a huge deal! My work is unrecognizable from what it was even 2 years ago. But that's an impact / productivity argument, not an argument about intelligence. Modern programming languages, IDEs, spreadsheets, etc... also made a fundamental shift in what being a software engineer was like, but they were not generally intelligent.
I do want to push back on some things:
> We treat "cognitive primitives" like object constancy and causality as if they are mystical, hardwired biological modules, but they are essentially just
I don't feel like I treated them as mystical - I cite several studies that define what they are and correlate them to certain structures in the brain that have developed millennia ago. I agree that ultimately they are "just" fitting to patterns in data, but the patterns they fit are really useful, and were fundamental to human intelligence.
My point is that these cognitive primitives are very much useful for reasoning, and especially the sort of reasoning that would allow us to call an intelligence general in any meaningful way.
> This "all-at-once" calculation of relationships is fundamentally more powerful than the biological need to loop signals until they stabilize into a "thought."
The argument I cite is from complexity theory. It's proof that feed-forward networks are mathematically incapable of representing certain kinds of algorithms.
> Furthermore, the obsession with "fragility"—where a model solves quantum mechanics but fails a child’s riddle—is a red herring.
AGI can solve quantum mechanics problems, but verifying that those solutions are correct still (currently) falls to humans. For the time being, we are the only ones who possess the robustness of reasoning we can rely on, and it is exactly because of this that fragility matters!
> who feed their produced tokens back as inputs, and whose tuning effectively rewards it for doing this skillfully
Ah, this is a great point, and not something that I considered. I agree that the token feedback does change the complexity, and it seems that there's even a paper by the same authors about this very thing! https://arxiv.org/abs/2310.07923
I'll have to think on how that changes things. I think it does take the wind out of the architecture argument as it's currently stated, or at least makes it a lot more challenging. I'll consider myself a victim of media hype on this, as I was pretty sold on this line of argument after reading this article https://www.wired.com/story/ai-agents-math-doesnt-add-up/ and the paper https://arxiv.org/pdf/2507.07505 ... who brush this off with:
>Can the additional think tokens provide the necessary complexity to correctly solve a problem of higher complexity? We don't believe so, for two fundamental reasons: one that the base operation in these reasoning LLMs still carries the complexity discussed above, and the computation needed to correctly carry out that very step can be one of a higher complexity (ref our examples above), and secondly, the token budget for reasoning steps is far smaller than what would be necessary to carry out many complex tasks.
In hindsight, this doesn't really address the challenge.
My immediate next thought is - even solutions up to P can be represented within the model / CoT, do we actually feel like we are moving towards generalized solutions, or that the solution space is navigable through reinforcement learning? I'm genuinely not sure about where I stand on this.
> I don't have an opinion on this, but I'd like to hear more about this take.
I'll think about it and write some more on this.
But, there's the time to set all of this up (which admittedly is a one-time investment and would amortize).
And there's the risk of having made a mistake in your backups or recovery system (Will you exercise it? Will you continue to regularly exercise it?).
And they're a 3-person team... is it really worth your limited time/capacity to do this, rather than do something that's likely to attract $3k/mo of new business?
If the folks who wrote the blog see this, please share how much time (how many devs, how many weeks) this took to set up, and how the ongoing maintenance burden shapes up.
This was pretty interesting to implement!
- I used an [lsp server](https://github.com/dlants/magenta.nvim/pull/162/files#diff-3...) to track opened files and aggregate text changes to get a stream of diffs.
- I then feed that along with the [context surrounding the cursor](https://github.com/dlants/magenta.nvim/pull/162/files#diff-1...), and a [system prompt](https://github.com/dlants/magenta.nvim/pull/162/files#diff-a...) into an LLM, [forcing a tool use for a find/replace within the context window](https://github.com/dlants/magenta.nvim/pull/162/files#diff-1...)
- Finally, I show the find/replace in the buffer using [virtual text extmarks](https://github.com/dlants/magenta.nvim/pull/162/files#diff-1...), applying a comment effect to the added sections, and a strikethrough to the removed sections
One thing that is interesting about this is that I wasn't able to get good results from smaller/faster models like claude haiku, so I opted to use a larger model instead. I found that the small delay of about a second was worth it for more consistent results.
I also opted to have this be manually triggered (Shift-Ctrl-l by default in insert or normal mode). This is a lot [less distracting](https://unstable.systems/@sop/114898566686215926).
One cool thing is that you can use a plugin parameter, or a project-level parameter to [append to the system prompt](https://github.com/dlants/magenta.nvim/blob/main/node/option...). I think by providing additional [examples of how you want it to behave](https://github.com/dlants/magenta.nvim/blob/main/node/provid...), you can have it be a lot more useful for your specific use-case.
I decided to record a video of myself using the plugin to implement another feature. Aside from a demo of the plugin itself, I think it's a good view into what a typical workflow of using AI for development might look like, and I think is a good illustration of what these agents can and cannot currently do.
Find out more here: https://github.com/dlants/magenta.nvim
I think the question is - why is the national guard and the military being sent to CA without the governor's consent?
Part of the justification for this from the administration is that the riots are out of control, are posing an immediate risk of violence and property damage. Based on what I've found of the actual violence and damage being done, this justification does not hold up, as the violence and property damage are lower than previous protests in which there didn't seem to be a need for interference.
The rule of law is a different argument. What is the rule of law that is being undermined? I think here too you can have an argument about the operations that ICE is conducting, are they lawful - given that they are being conducted in sanctuary cities? Who has jurisdiction in this case? Is the administration lawful in sending in the national guard without the consent of the governor? What about the military? Newsom is now suing Trump over deploying the marines and the national guard despite his wishes, so there is a claim that such actions were unlawful.
Based on my brief research into this, ICE was operating in an unconstitutional way and making many procedural violations. City Sancutary status is lawful and has been upheld in the court of law. Newsom's challenge of Trump's deployment of the National Guard also held up in court. [link](https://claude.ai/public/artifacts/99a64b2d-e3b6-4d37-956f-c...)
When you say that "the law is being attacked" in this case, what do you mean?
How extensive is the violence of the protests? I saw some images shared of cars that were burned, maybe some buildings damaged. But also lots of images from other protests from previous years. Are the images of the same 3 cars and storefronts or many? Trump says the riots are out of control, Newsom says the protests are largely peaceful.
A basic claude search suggests the overall level of violence is moderate, and smaller than many recent protests [link](https://claude.ai/public/artifacts/ef220c3d-c6d9-4b4b-bb3f-2...)
How much of a strain do undocumented immigrants place on the US? You can answer this question from a financial and criminal point of view. From the point of view of crime, Trump and ICE are parading every violent undocumented immigrant they can, but that is not statistics. Do undocumented immigrants account for a significant portion of violent crime in the country?
Studies overwhelmingly show that undocumented immigrants are significantly less violent than the general population [link](https://claude.ai/public/artifacts/a92623b8-5c02-4c3a-84ae-f...)
From a financial point of view, what resources are undocumented immigrants straining, and is it to a significant degree?
The economic picture is much more nuanced. On the cost side, a criticized study (FAIR) reported the cost at about $182bn annually (this is likely an over-estimate). For comparison, undocumented immigrants pay about $100bn in taxes, boost the GDP, and create jobs. Mass deportation is estimated to cost $315Bn.
Studies show that the impact on wages is small.
The biggest cost factor ($78bn but estimates vary) seems to be K-12 education, and that is mostly born by states. [link](https://claude.ai/public/artifacts/29f10fcf-c8a7-4655-979f-b...).
In using it I've found that the agent doesn't really make use of these tools unless I instruct it to. I think I need to do some messing around with prompts to encourage the agent to use them more.
I think the next major thing I want to add is commands to allow you to send symbol types / diagnostics to the agent before it asks for them, which should help speed up some of these workflows.
Thanks for the pointer to take another look at aider - the DIFF format sounds really interesting. Though it seems like the license is Apache 2.0 so I think I'll have to learn a bit about what that means since I made mine MIT licensed.