mediaman

5 ms·

1.
by mediaman 6d ago
Agreed. I find it really fun to work with customers. It keeps you focused on real problems. And there are so many problems in 'boring' industries that need solving. Just a ridiculous number. There's so much bad software out t
2.
by mediaman 6d ago
This is a groundless criticism. TB2.1 is saturated. TB4 is not. Sol xhigh is 90% on TB2.1 but 37% on TB4. Is it also "benchmaxxed"? Your assumption is that the benchmarks are essentially identical in difficulty, with the only diff
3.
by mediaman 20d ago
No, that's not correct for any reasonable definition of "impossible." Look up pangram's accuracy ratings. It's not perfect, but it's pretty good. LLMs in fact leave very distinguishing traces of their logit dis
4.
by mediaman 21d ago
Just look at the prices that inference providers charge for small models. The argument that these unit economics are negative is trivial to disprove. DeepInfra sells DS v4-flash at 0.08 in, $0.18 out. Gemma4 they sell for $0.07 in, $0.34 ou
5.
by mediaman 21d ago
You can make the other argument that China subsidizes the price and that they can't be profitable at this pricing level. From an industrial strategy standpoint, they already do this for many other industries with huge subsidized state
6.
by mediaman 26d ago
Overall I prefer Codex over Claude in their current state, but I do agree that Codex's solutions tend to be too complex and I have to constantly tell it to simplify solutions, detect when it's creating ornate architecture, and pro
7.
by mediaman 26d ago
As a coda to this, anyone using grok 4.6 via API pricing should be aware that while their headline pricing is good, the pricing that actually matters is pretty bad. Their cache read costs are $0.50 per million, or 25% of the cost of uncache
8.
by mediaman 26d ago
Criminal acts do not require the victim to "press charges." A government prosecuting attorney decides whether to criminally prosecute the alleged perpetrator. "Pressing charges" is mostly a made up idea for criminal case
9.
by mediaman 29d ago
Consider that the stance now is banning data centers . That's the context of the discussion. The county's other option is to simply price data centers . Yes, that means they might not win the data center. If the data center can
10.
by mediaman 29d ago
My theory is demand substitution. NVME/SSD is up so much people at the margin are switching to spinning rust, which nobody was prepared for, so now they too are drowning in demand.
11.
by mediaman 29d ago
Simple mechanistic explanations in biology are often wrong. For example, oxidative stress does not have a linear no-threshold harm. For alcohol, almost every big study has shown either no discernible effects going from zero to 1 drink per d
12.
by mediaman 29d ago
That's not a study from the WHO, it's a public statement they published in Lancet. It's an opinion piece. Moreover, the WHO never changed their stance when later big studies showed that the "lowest risk" level is no
13.
by mediaman 29d ago
What I don't understand is why communities want to block the data centers. Just make them pay! Your electricity price shouldn't go up because of them. So just make a rate card for data centers that ensures the burden doesn't
14.
by mediaman 29d ago
To continue to take us off topic, there is actually pretty limited evidence that this is true. It is true that wine doesn't do anyone any good (unless you count it making people more social or something). But the effects of one glass a
15.
by mediaman 1mo ago
Clickhouse has a more intentionally built ingestion system. Duckdb has concurrency limits so you can't have a writer and a reader on the same file if they're not the same process (multiple readers is fine). But that's not too
16.
by mediaman 1mo ago
It's possible, but I've experimented with different style guides and prompt methods and have never gotten it to output what I'd call "good" writing. It is almost always low information density. Sometimes I can get i
17.
by mediaman 1mo ago
I do something similar and just use echarts. Very happy with it.
18.
by mediaman 1mo ago
I built a platform for some midsize companies in a specific vertical that is basically a data warehouse with some LLM-driven dashboarding and query tools on top. Typical data size 5-150gb. So I built a service layer around duckdb, where eac
19.
by mediaman 1mo ago
That poster mixed it up with trademarks, for which enforcement is required to maintain its validity.
20.
by mediaman 2mo ago
I don't see how this follows. The cost of nails has fallen by 95% over the last century. It's because the cost of manufacturing has fallen. Not because they are selling the information of nail consumers. Tokens are not normal soft
21.
by mediaman 2mo ago
Yes but there is a big, big market for subagents to consume lots of tokens cheaply and condense information up to parent agents. Luna would not be my choice for planning. But an explorer to comb through a codebase to find relevant parts? Or
22.
by mediaman 2mo ago
So this only switches models if the cache is cold, because otherwise the economics of switching don't work. But most agentic work involves long strings of successive tool calls that benefit from a hot cache. Hot cache calls reduce inpu
23.
by mediaman 2mo ago
None of this matters, this is the judge approving a voluntary settlement reached between the parties last year. If you think it should be different then you have to make a cogent argument why the public should get to interfere with a settle
24.
by mediaman 2mo ago
He's saying they just plug your USB stick into some gizmo sold to the state, and it's going to find enough to escalate it. If the idea is that there's no point in encrypting anything on a USB stick, and you just hope they don
25.
by mediaman 2mo ago
Because they don't have a lot of parameters to store general Wikipedia knowledge. They're small. Use big models that have high parameter capacity to store general information. Or build a harness around the small model that searche
26.
by mediaman 2mo ago
Small open source models shouldn't be used for world knowledge, that's not their purpose.
27.
by mediaman 2mo ago
Gemma 4 was released in April. It's a good series of multimodal models.
28.
by mediaman 2mo ago
There was some recent reporting that a July release of the Pro model got pushed back for exactly that reason. Its performance was not good compared to the OpenAI/Anthropic big models. They are having a lot of problems with posttrain.
29.
by mediaman 2mo ago
You're correct, but that's a different market segment and not the market GLM 5.2 and its peers compete in. The labs are not interested in the small, fast, single purpose end of the market. Google increased their pricing on Flash s
30.
by mediaman 2mo ago
The (quite excellent) article discusses several of your points. If you haven't read it, I recommend it. - Commodity market profitability is determined by marginal cost of production. LLMs have marginal cost; traditional software do

More ›