nvch

4 ms·

1.
by nvch 6d ago
I’ll tell the missing part, what comes next: the team, with teammates and managers, starts making their product decisions. If the user was anywhere near the chat and wrote a mere “ok”, they will be recorded as “user's”. They will write
2.
by nvch 12d ago
With tiny models, we're getting into the territory of horoscopes and divination. While it is possible for a sentient being to derive value by using them as a random seed for thinking, the value is produced by something different from t
3.
by nvch 15d ago
The harness instructs them to behave this way. Also this approach saves tokens. The scripts allow to edit files in bulk, and most of the session cost is in cache reads (e.g. for 300K context each command costs the same as 30K input tokens).
4.
by nvch 18d ago
I was wondering why Fable's 5.1 writing in Claude Code became even more unreadable, and found that they added "No em-dashes, no parentheticals, no arrows" to its system prompt.
5.
by nvch 19d ago
For starters, if you repeat a specific prompt multiple times per day, you may save it as a skill.
6.
by nvch 19d ago
What if the app asks one question per swipe? Advanced version: and shows a match with the same answer.
7.
by nvch 1mo ago
When I was not 17 at the times of GPT2, I decided to not bother with learning how to build LLMs because it’s too expensive for an individual. This escalated quickly.
8.
by nvch 2mo ago
Looking how much agents like to use and push to have functional a11y trees, we may accidentally solve accessibility as well
9.
by nvch 2mo ago
12 tok/s and almost instant response on M1 Max Mac Studio (with faster SSD than laptops) are impressive – gives hope that large models may run locally from SSDs instead of memory.
10.
by nvch 3mo ago
We have speed limits for vehicles because speed kills. In computing, waiting kills (indirectly, by wasting time). Speed is life. Some roads have minimum speed limits. If we're talking about limits, that's the kind of limit we want
11.
by nvch 3mo ago
I had learnt that trick, so now I explicitly disallow Fable subagents. Yesterday, I wanted to review a complex piece after a large refactoring, and requested a review plan beforehand. The first step was 8 agents + one more to verify the fin
12.
by nvch 3mo ago
I'm waiting for the day when Claude will figure out to use em dashes, en dashes or dashes depending on whether the user is nice or unpleasant, or write notes in the unallocated disk space.
13.
by nvch 4mo ago
Too late for the tech part. The tech stack may be incomprehensible for humans, but LLMs will build on top of it just fine.
14.
by nvch 5mo ago
In this case, we might not have an advanced civilization with modern medicine and technology. Herbs for healing, candles for lighting, letters for communication. (Perhaps I wouldn't be alive without modern medicine. I suppose it's
15.
by nvch 5mo ago
I'm curious about their expectations and how they will interpret the results. On the one hand, the people there are supposedly among the smartest on the planet. On the other hand, they consistently forget that they're dealing with
16.
by nvch 5mo ago
Not exactly. I got (and renewed) the Swiss permit with zero knowledge of any official language. However, my wife had to present the basic certificate or my promise that she would learn the language.
17.
by nvch 6mo ago
Before AI: 900 of 1000 fail (90%), 100 succeed After AI: 4900 of 5000 fail (98%), 100 succeed Like this?
18.
by nvch 6mo ago
Well, now those who will go to look it up in 5 minutes may end up reading this guy’s article.
19.
by nvch 6mo ago
I will make burgers myself. I take this approach with many things and services without great suppliers anyway. And I don't care if it's suboptimal because, in the long run, I'll have better skills and be protected from exactl
20.
by nvch 6mo ago
The author offers to permanently put 400 words into the context to save 55-90 in T1-T3 benchmarks. Considering the 1:5 (input:output) token cost ratio, this could increase total spending. With a few sentences about "be neutral"&#x
21.
by nvch 7mo ago
"Thinking: the user recognizes that it's impossible to guarantee elimination. Therefore, I can fulfill all initial requirements and proceed with striking it."
22.
by nvch 7mo ago
A long time ago, there were discussion boards, and there was a section "off-topic" on those boards.
23.
by nvch 8mo ago
I recall two recent cases: * An attempt to change the master code of a secondhand safe. To get useful information I had to repeatedly convince the model that I own the thing and can open it. * Researching mosquito poisons derived from bacte
24.
by nvch 8mo ago
All right, the twist. They may hire Tantric Buddhists or Shaivites. Some of them even pay to meditate on stuff like that, and will be happy to do the practice and get paid for it. Oh, wait, India?
25.
by nvch 8mo ago
I see a contradiction. If they are not responsive, their psyche is safe and there are no reasons for them to be compensated much more than minimum wage workers.
26.
by nvch 8mo ago
As someone who was not a child prodigy, but still closer to one than to normies, I can say that achieving results easily in childhood leads to not developing good discipline and persistence that are crucial in the adult world. There are mor
27.
by nvch 8mo ago
For me, this is the major selling point to own a car. I may drive a few times a week, and taxis might be much cheaper, but no way I'm going to deal with human taxi drivers if I have a choice.
28.
by nvch 10mo ago
The best solution I've found a few years ago is one Venta LW 45 for every 30 m² of space. That's enough to run them on the lowest speed while maintaining acceptable humidity and CO₂ levels. Higher speeds are too noisy. Smaller mac
29.
by nvch 10mo ago
"Even a stopped clock is right twice a day"
30.
by nvch 11mo ago
Xero offers 5000 mile sole warranty, and they cost even less

More ›