thomasahle

8 ms·

91.
by thomasahle 11mo ago
It's only the later if it's a weak browser engine, and it's early enough in the game that the player had studied the position with a cloud engine.
92.
by thomasahle 11mo ago
I also used Gemini 3 Pro Preview. It finished it 271s = 4m31s. Sadly, the answer was wrong. It also returned 8 "sources", like stackexchange.com, youtube.com, mpmath.org, ncert.nic.in, and kangaroo.org.pk, even though I specifical
93.
by thomasahle 11mo ago
I tried it with gpt-5.1 thinking, and it just searched and found a solution online :p
94.
by thomasahle 11mo ago
Even if you personally had a mathjax extension, you would still be prevented from explaining math to others, unless you could convince everyone to install it.
95.
by thomasahle 11mo ago
How would you write this with matrices? It seems like there are many ways you could generalize.
96.
by thomasahle 11mo ago
There are lots of good books on writing mathematics: 1. How to Write Mathematics — Paul Halmos 2. Mathematical Writing — Donald Knuth, Tracy Larrabee, and Paul Roberts 3. Handbook of Writing for the Mathematical Sciences — Nicholas J. Higha
97.
by thomasahle 11mo ago
What a weird way to write the harmonic average. ---- Write v_i = Var[X_i]. John writes t_i = \frac{\prod_{j\ne i} v_j}{\sum_{k=1}^n \prod_{j\ne k} v_j}. But if you multiply top and bottom by (1 / \prod_{m=1}^n v_m), you just g
98.
by thomasahle 11mo ago
Github: https://github.com/thomasahle/trace-taxi Link: https://trace.taxi
99.
Show HN: Trace.taxi – easy agent messages visualization
3 points by thomasahle 11mo ago|1 comments
100.
by thomasahle 11mo ago
I expected this to be about some "AI first" terminal. But the article doesn't mention AI at all, for better or worse...
101.
by thomasahle 11mo ago
For Senior roles!
102.
by thomasahle 11mo ago
Normal Computing | Multiple Roles | New York, San Francisco, London, Copenhagen | ONSITE/HYBRID | Full-time | normalcomputing.ai Normal Computing is building a new thermodynamic computing paradigm to revolutionize AI reasoning and reli
103.
by thomasahle 11mo ago
I doubt it, since jaxtyping supports some quite advanced stuff: def full(size: int, fill: float) -> Float[Array, "{size}"]: return jax.numpy.full((size,), fill) class SomeClass: some_value = 5
104.
by thomasahle 11mo ago
Some earlier discussions (2022): https://news.ycombinator.com/item?id=31441979 Most critique of KaTeX over MathJax is reduced support for LaTeX features.
105.
by thomasahle 11mo ago
Ah, I would have never thought jaxtyping supports torch :)
106.
by thomasahle 11mo ago
You can do that in python using https://github.com/patrick-kidger/torchtyping looks like this: def batch_outer_product(x: TensorType["batch", "x_channels"], y:
107.
by thomasahle 1y ago
> text tokens are discrete while image tokens are continuous. Each model has a finite number of text tokens - say, around 50,000. Each of those tokens corresponds to an embedding of, say, 1000 floating-point numbers. Text tokens thus onl
108.
by thomasahle 1y ago
You are describing the same thing, but at different levels of explanation Llamasushi's explanation is "mechanistic / representational", while yours is "behavioral / statistical". If we have a pipeline: `tr
109.
by thomasahle 1y ago
> Category theory can be used to express some requirements in a very concise way. Can you an example?
110.
by thomasahle 1y ago
> I miss the time when computers did what you told them to, instead of trying to read your mind. You haven't seen anything yet. Get ready for "Sort by AI" which will try to interpret the content of your images to sort them
111.
by thomasahle 1y ago
> While a democratic state is different than a private organization in that it derives legitimacy from its voters, I'd be hard pressed to say that the state is sufficiently different from any other large organization. If the voting
112.
by thomasahle 1y ago
> The duo kept their program running in the background for over a decade Were they not searching for counter examples to the sum conjecture during this time? Or how did they program not identify this simple 3+3 example sooner?
113.
by thomasahle 1y ago
Second answer is better imo: > Don't use My or Your. In most cases it's obvious whose they are. > The only case you might want to do it is to differentiate e.g. between the user's documents and everyone's documents
114.
by thomasahle 1y ago
Interview: > We can solve this with a constraint solver Ok, using your favorite constraint solver, please write a solution for this. > [half an hour later] Ok, now how would you solve it if there was more than 100 data points? E.g. 10
115.
by thomasahle 1y ago
Comparing to Haskell, I think of "async" as the IO monad. It's nice to have all code that does IO flagged explicitly as such.
116.
by thomasahle 1y ago
> > Doing work that matters might. > This is a pre-requisite for winning on LinkedIn. The kind of content that performs best are strong opinions informed by actual expertise. > > Go for depth over frequency. > Unfortunatel
117.
by thomasahle 1y ago
That's fine if you use it for private use. Doesn't work if you're building a product using Claude.
118.
by thomasahle 1y ago
Other than entirely community-driven projects (like https://wanderer.to/ mentioned in the article), are there company "forms" that legally protect against this kind of sell-out? Like non-profit or public-benefit-c
119.
by thomasahle 1y ago
> But the Tower of Hanoi can be solved without "tools" by humans, simply by understanding the problem, thinking about the solution, and writing it out. The paper doesn't give any evidence humans are able to do this. And I
120.
by thomasahle 1y ago
Exactly. If the paper title had been "LLMs are not that great at thinking", nobody would have had an issue.

More ›