HN user
False-positive rate is so high with Mythos according to friends and other reporting I have seen.
The original Mythos release used ASan to filter false-positives so it was able to maintain a good FPR, but when Mythos moves into domains that don't have a readily available oracle to help filter hits, the result is a deluge of false bullshit.
Newport's book on this topic is terribly underwhelming, and I say that as someone who has really enjoyed his other books, blog posts, and YouTube videos. Most of the anecdotes he gives to support his "Career Capital over Passion" have an ambiguous directionality of the causal arrow. For example, the fact that the happiest admin assistants have been at the job the longest does not mean that getting good at being an AA makes you happy with the work. There is an equally plausible explanation that enjoying the work of an AA makes one likely to stay in the job longer. Most of his examples in the book are like this.
The place where I think Newport flounders in this area is that, in order to get "So Good They Can't Ignore You", you actually have to be able to put in the time and effort to get good. And the vast majority of people do not possess the self-control and willpower to force themselves to do something they dislike to the level that is necessary to achieve said mastery.
Yeah, LLMs suck at named lifetimes. The number of times I have seen Claude reach for indices and clones instead of just using proper named lifetimes is too many to count at this point. Not great for high-performance code!
No, you don't understand. LLMs will never be capable of knowing what questions to ask, only how to ask the questions. /s
It would be helpful to add in some cases that do not contain any vulnerabilities to assess false-positive rate as well.
In my personal experience, the rate at which Claude Code produces suboptimal Rust is way higher than 1%.
But Eon's tagline is "Solving brain emulation as an engineering sprint, not a decades-long research program"! How could they have ever gone wrong?
I especially love the quadratic fit, chosen with no justification, that brings the US within the uncertainty envelope in the second plot. Also notice how much work the Mexico and USA data points are doing to the previous linear model fit. Oh, my high leverage data point can't be an outlier because it's within uncertainty when I fit the data with the potential outlier included. This is basic linear model validation stuff.
Has a hard to explain fixation on doing things a certain way, e.g. always wants to use panics on errors (panic!, unreachable!, .expect etc) or wants to do type erasure with Box<dyn Any> as if that was the most idiomatic and desirable way of doing things
Yes! I see this constantly. I have a Rust guide that Claude adheres to maybe 50% of the time. It also loves to allocate despite my guide having a whole section about different ways to avoid allocations.
Two use-cases recently where Claude sucked for me:
1. Performance-critical code to featurize byte slices for use in a ML model. Claude kept trying to take multiple passes over the slice when the featurization can obviously be done in one. After I finally got it to do the featurization in one pass it was double-counting some bytes but not others (double counting all of them would have been fine since the feature vector gets normalized). Overall it was just very frustrating because this should have been straight-forward and instead it was dogshit.
2. Performance-critical code that iterates over lines of text and possibly applies transformations, similar to sed. Claude kept trying to allocate new Strings inside of the hot-loop for lines that were not transformed. When I told it to use Cow<'a, str> instead so that the untransformed lines, which make up the majority of processed lines, would not need a new allocation, Claude completely fucked up the named lifetimes. Importantly, my CLAUDE.md already tells Claude to use copy-on-write types to reduce allocations whenever possible. The agent just ignored it, which is _the_ issue with LLMs: they're non-deterministic and any guidance you provide is ultimately just a suggestion.
It's not just assuming that everyone learns the same way. It's assuming that everyone learns the way that all of the research literature on learning claims does not work.
Learning requires active recall/synthesis. Looking at solved examples instead of working them yourself does not suffice in math, physics, chemistry, or CS, but somehow it is supposed to work in this situation?
Machine learning people use "tensor" to just mean an N-dimensional array of numbers. The term is divorced from its meaning in Physics and Mathematics, which caused me some confusion when I started looked at machine learning papers coming from physics.
But the energy transported to Earth from your space power plant still creates waste heat when it is used to do work (and also when it is transported to earth). You cannot beat the second law.
They don't even decouple at high material standards of living. Recent increases to GDP produced emissions too but those new emissions were offset by reductions in emissions of existing industries.
This "decoupling" gets us basically nothing because it's not like we can just stop emissions tomorrow since GDP and emissions are "decoupled".
How does UDP work if you're also using delta compression? I would naively expect that the accumulation of lost diff packets over time would cause game state drift among the clients.
It's $100M to develop tech that would allow carbon to be pulled out of the atmosphere at the price of $100-$500 per ton of CO2. Even if the tech is successful then it would still cost $100B-$500B per year to pull out the 1 bilion tons of CO2 by 2050 that the IPCC has built into their models.
The tech for Direct Air Capture (DAC) is already quite efficient at removing CO2 from the atmosphere (80%+ depending on the particular process). At best, new innovation on the chemistry of the removal process can only increase the efficiency by 25%.
The real issue with DAC is that it is incredibly difficult to innovate around the fact that CO2 in air is just immensely dilute. You need to process enormous amounts of air to remove an appreciable amount of CO2 and, even worse, as the plant operates and recirculates processed air, the local air around the plant becomes more and more devoid of CO2 leading to a decreased amount of CO2 captured per unit volume of air.
The only real improvement that I could see occurring in this space is coming up with a process that creates less back-pressure against the fans pumping in unprocessed air which could bring down the energy cost per ton of CO2 removed. But even in that case the back pressure created by a DAC process is normally caused by flowing air through a porous catalyst, which is essential for high efficiency. So there's a trade-off there as well.
Ultimately I am not very sanguine about DAC and I have been disappointed to see news agencies reporting as if DAC is even in the top 10 technologies most important to reduce carbon emissions.
It is wrong to frame the argument as wind/solar versus everything else. The argument should be carbon versus no carbon. Period.
I am framing it as carbon vs no carbon. A new nuclear plant's timeline to carbon-neutrality compared to the counterfactual where an equal capital investment is made in solar/wind is well over a decade. Given the time crunch we are on to lower emissions, I simply do not think we have the time to waste building new plants. That said, we should not be decommissioning plants that are still operable.
This is the best comment in the thread and it's somehow being downvoted.
Physics grad student here and I agree with this comment whole-heartedly. I love the idea of nuclear power but I also understand that it requires enormous CapEx and that the time to get a reactor up, running, and carbon neutral is much too long to address the climate crisis. We absolutely shouldn't be shutting down nuclear plants but any money spent on new plants is money that could otherwise be spent generating lower cost Wind/Solar in a shorter period of time.
Why are you complaining that the author didn't talk about tensors as they are used in tensorflow? Tensorflow is never even mentioned in the piece.
The author is perfectly clear in the first sentence that the piece's focus is about the usefulness of tensors in a physics context.
Replying to my own comment with an update.
The temperature today in my area varied between 9 F at 9AM and 25 F at 3PM. My girlfriend happened to need to drive to several different towns/cities for her work so her total trip was 135 miles. Our Bolt says it has 50 miles left.
I only charge to 90% and just received a new 259 mile range battery as part of the Bolt recall. So the EPA range of my Bolt at 90% charge is 233 miles. If we take the Bolt's 50 mile left range claim at face value (which I have found to be overly pessimistic), then it got 185 miles of range today out of the expected 233, or a ~20% drop in EPA stated range.
So I just don't believe the numbers provided in the linked article. They are completely out of whack with my every day experience owning a Bolt through three winters in Maine.
I have to question the accuracy of the numbers presented in this article.
I own a Chevy Bolt in Maine and the figure listed in this article is far off from what I have experienced over several winters here. At a temp of 20 F my bolt loses ~20% of its range from the EPA estimate, going from 236 to about 190miles of range. Conversely, my friend owns a Tesla Model 3 up here and his range degrades severely during winter on the order of 30%.
edit: He owns a Model 3, not a Model Y.
This same class did the same thing to me at Princeton except I was an engineering major at the time. I had done well on the Calc AP exams in high school which the engineering department said placed me in MAT 202. On the first day of class the professor started summarizing "what we already knew" from high school about linear algebra and I had seen basically none of it with the exception of basic matrix multiplication. The rest of the week was ego-destroying as I attempted to get help from the professor and was repeatedly told (almost berated) that I should already know the answers to my questions from my high school courses.
I eventually dropped the course, left engineering entirely, and majored in a biological science. But the jokes on them because I self-studied a shit ton of math after I graduated and eventually went back to grad school doing ML+Physics. It turns out that I'm actually pretty good at Linear Algebra after all.
Due to dietary restrictions my basket has remained mostly unchanged for years. Every grocery trip is basically vegetables, meat, brown rice, milk, cheese, and cooking oils or spices if needed.
Just added it.
I keep a monthly accounting of my grocery costs (along with all other expenses) in a spreadsheet. I haven't seen my grocery bill increase by anywhere close to 40% over the past two years.
edit: My bill went from ~$600 to ~$650 using a 4 month window rolling average.
Direct Air Capture (DAC) will never be impactful for the simple reason that you need to process way too much air to remove an appreciable amount of Carbon. Capture at the source of plants producing CO2 will be helpful, but DAC is a pipe dream.
They are using numbers for a fucking fun size snickers, ie the tiny ones people give out at Halloween that are a single bite.
If you aren't a landlord then why would you think the comment was referring to you?
I do and I do.