mediaman

9 ms·

91.
by mediaman 6mo ago
It's surprising to me how much LLM "personality" seems to matter to people, more than actual capability. I do turn to Anthropic for ideation and non-tech things. But I find little reason to use it over codex for engineering t
92.
by mediaman 7mo ago
It's not that expensive. The Starlink Mini is around $200, and service is $50/mo for 100gb. I've been somewhat skeptical of the addressable market (doesn't fiber + cell tower network offer good enough coverage?) but I kn
93.
by mediaman 7mo ago
I'm unclear what statement this is trying to make. Is it meant to draw equivalence between crypto and Tesla/SpaceX? That each has roughly similar (i.e., low) value to humanity, or value as businesses? Is it that the metric of whet
94.
by mediaman 8mo ago
10 USC 3252 has only been used once, against Acronis AG, a Swiss company with Russian connections. Acronis did not have DOD contracts. Other companies (Huawei) have been deemed risks under different laws, or by Congress, but they also didn&
95.
by mediaman 8mo ago
It's odd because I no longer really like ChatGPT. For chat-type requests, I prefer Claude, or if it's knowledge-intensive then Gemini 3 Pro (which is better for history, old novels, etc). But GPT 5.3 Codex is great. Significantly
96.
by mediaman 8mo ago
Why not? It's a physical building with lots of equipment that produces products shipped to its customers. Its products are sequences of electrons, instead of atoms. But so are power plants. And in the context of what happens when they&
97.
by mediaman 8mo ago
They have not; a social media post does not satisfy the requirements of 10 USC section 3252. They are required to notify Congress (they have not), prepare a report with specific sections (they have not), and the reasons must fall within a s
98.
by mediaman 8mo ago
Mass domestic surveillance of American citizens (they were OK with surveillance of other countries).
99.
by mediaman 8mo ago
It's not either of those. Anthropic put a lot of effort into getting FedRAMP approved so the DOD could use them; they are now being punished for that, and the government at present has no other good options. Other options could of cour
100.
by mediaman 8mo ago
The product is a service, and they agreed to a contract. Now they don't like the contract. Is your view that contracts with the government should be meaningless? That the government should be able to unilaterally, and without recourse,
101.
by mediaman 8mo ago
The 40k lines of code a day crows are amusing. In solving any problem solvable by code, there's a ratio of non-coding work to coding work, and codex et al all help immensely with the coding work but help less with the non-coding work.
102.
by mediaman 8mo ago
Others clarified the kW versus kWh, but to re-visit the comparison to a household: One household uses about 30 kWh per day. 20 kW * 24 = 480 kWh per day for the server. So you're looking at one server (if parent's 20kW number is a
103.
by mediaman 8mo ago
People from the West always complain when people from the developing world are hired to do work for people in the west. They got angry about China, the Philippines, India, Kenya. Oddly, it’s never the people in those countries complaining t
104.
by mediaman 8mo ago
If there are low switching costs, and if there are multiple highly capable models, and if the hardware is openly purchasable (all of these are true), then the price will converge to a reasonable cash flow return on GPUs deployed net of oper
105.
by mediaman 8mo ago
The comment was with reference to inference, not total P&L. Of course they are losing money in total. They are not, however, losing money per marginal token. It’s trivial to see this by looking at the market clearing price of advanced o
106.
by mediaman 8mo ago
Seems trivial to create an infinite number of inconsequentially (but hash defeating) different variants.
107.
by mediaman 8mo ago
Africa, Europe, America, Mars. I wonder if there is something about one of these that makes them unlike the others. Actually, why not colonize Venus instead? Sure, it will be hard, at first, with all the sulphuric acid and intense heat and
108.
by mediaman 9mo ago
What does it mean, that open world winning was a mistake? That the market is wrong, and peoples' preferences were incorrect, and they should prefer small handcrafted environments instead of what they seem to actually buy?
109.
by mediaman 9mo ago
How? They are all losing tens of billions of dollars on this, so far. Open source models are available at highly competitive prices for anyone to use and are closing the gap to 6-8 months from frontier proprietary models. There doesn't
110.
by mediaman 9mo ago
I think much of the rot in FAANG is more organizational than about LLMs. They got a lot bigger, headcount-wise, in 2020-2023. Ultimately I doubt LLMs have much of an impact on code quality either way compared to the increased coordination c
111.
by mediaman 9mo ago
Some are complaining this letter is weak and generic. Of course it is. You have 3M, Target, General Mills, Cargill, and US Bancorp on here, among others. If you are looking for some revolutionary call to action, you're looking in the w
112.
by mediaman 9mo ago
It is, in fact, not crazy, because none of this is predicated on using a specific vendor. Many of these techniques can also work with Chinese LLMs like Qwen served by your inference provider of choice. It's about the harness that they
113.
by mediaman 9mo ago
I don't get the widespread hatred of Gas Town. If you read Steve's writeup, it's clear that this is a big fun experiment. It pushes and crosses boundaries, it is a mixture of technology and art, it is provocative. It takes st
114.
by mediaman 9mo ago
Yes - I've been thinking about why this is. I'm guessing part of it is that writing forces us to think. I often find when I write something that I haven't thought it out fully, and articulating it makes me see a logical failu
115.
by mediaman 9mo ago
Survivor bias of those things that haven't been solved. Notably absent: The fat pill HIV fix Cystic fibrosis We make fun of the stuff that hasn't been solved yet ("It's always ten years away!") while ignoring the th
116.
by mediaman 9mo ago
No, they later updated the harness for this and it subsequently got better scores.
117.
by mediaman 10mo ago
RLVR is not offline learning. It's not learning from a static dataset. These are live rollouts that are being verified and which update the weights at each pass based on feedback from the environment. You might argue that traditional R
118.
by mediaman 10mo ago
I don't understand this. In supabase, the default is to turn on RLS for new tables. If you turn it on and have no policy set, no user can fetch anything from the table. You have to explicitly create a read-all policy for anon keys, and
119.
by mediaman 10mo ago
Much of this is due to vastly better posttraining RL, not models that are much bigger. The idea that most of these gains comes from training really big models, or throwing immensely larger amounts of compute at it, is not really true.
120.
by mediaman 10mo ago
This is a regurgitation of the old critique of history: what's it's purpose? What do you use it for? What is its application? One answer is that the study of history helps us understand that what we believe as "obviously corr

More ›