🌎 clearly my timing is off this week.
----
a plane went down last night. a military helicopter crashed into it.
additional details are still unclear. ⚙️ https://www.wusa9.com/article/news/special-reports/dc-plane-crash/all-flights-halted-at-reagan-national-airport-due-to-plane-crash-potomac-river-dc/65-e2090f2d-0bca-4a4c-944c-215a6398a52d
----
no follow-up thoughts on the machine that require repeating.
the Outstanding Question is: why do people feel so much better about the machine when it phrases its answer in a way that makes them feel smart? ⚔️ when you phrase it that way, the answer is kind-of obvious
the Second Outstanding Question is: what good things happen when you separate "instant chat" responses from "intelligent question-answering and task-processing" responses?
----
the question of can the machine participate in rituals is too sensitive to discuss in an open forum.
----
the question of can you give the machine a swiss-army knife and ask it to choose which tools to use is also uncertain. 💡 at some point, the answer will obviously be yes. but, does the "quick and dirty" approach work?
I find myself wanting to block Reddit, etc. on the router level. There is nothing worth reading there.
----
the real way to get an "a-ha" moment from the machine, is with time-travel. 🔥 with time travel, the machine can return a better answer instantly!
the "head" of the response has to be noticeably ahead of where "committed" responses are. so, there is a possibility to jump backwards.
🔥 isn't this beam-search?
💡 ... maybe.
but, the idea is: you can assert a token ERROR PATH: RETREAT 32.
then, the "reason" for the message can be added as input.
it is the a-ha moment. but, implemented better than DeepSeek.
----
the reaction to DeepSeek has been, in my estimation, ridiculous.
I tried the 7b and 8b distilled models. And what I saw was a cheap parody of thought. Thought-processes that didn't make sense, and didn't actually reflect how the machine generated thoughts.
but, apparently, people like it.
Maybe the 400b model gives better answers? Or, maybe, people just see the shape of the answer and trust it more.
💡 if the goal of the machine is to solve industrial tasks, this is mostly already baked in to my estimations. but, for the goal of making consumers happier, there is clearly a factor I am not considering.
----
theory 1: people don't want to think the machine is smart; they want the machine to make them feel smart. both the appearance of struggling and the visible chain-of-thought (even if obviously flawed) contribute to this feeling.
theory 2: people don't know that the machine could already do 90% of this 12 months ago. they see a demo (or, more likely, hear about a demo) and, miraculously, now they know what will happen.
theory 3: we know that a light human touch guiding the machine's responses can improve accuracy substantially. and, that human touch can also be automated.
⚔️ well, actually, apparently very few other people knew that.
----
i'm going to stick with Theory 1 for today. that people like Deepseek (and feel it is better) because it makes them feel smart.
which ... is depressing. but, also, easily solvable.
the question is: what question could you pose that would lead somebody to come up with this answer on their own?
💡 it seems unlikely that 8B models can do this. but I assume the 600B models can.
----
people want the machine to make them feel smarter. 🔥 because people are self-centered, gullible, and insecure.
💡 they want it to behave in a way that I instinctively hate. they want the PT Barnum version of AI.
💬 give the people what they want!
----
this is probably one of the reasons why the default tone for every chatbot is obsequious. so much that's a great question! / you're absolutely right / let me know what else i can do to help.
----
💡 one can apply a politeness filter to the output of the machine. but the latency of such a system is already high.
⚔️ well, actually, it probably is just another layer or two.
----
Seen on social media: Anthropic is losing because they have rate limits! 💡 of course they have rate limits. the machine is not too cheap to meter, at least at the quality people expect.
----
a game of chess.
the idea of the attack worked in theory. and the attack worked in practice. but the actual attack did not work, in theory.
✨ chess can be a ritual. like the i-Ching.
----
can the machine participate in rituals?
----
there are two kinds of answers people want from the machine.
🔥 perhaps testing is a third category.
Whereas: for many valuable use-cases, having a 5-minute latency to do it right, is not objectionable. 💡 the evocative questions, the "what do you mean by LONDON" and "can you talk more about LONDON", will be interactive. ⚙️ we do not have LONDON implemented here yet.
yesterday was a day for "rough drafts" (written elsewhere) and job applications. 🔥 a rough draft of my future!
today is a more normal day.
----
Atacama to-do
I keep considering "editing" / "publishing" features. But, I am yet to find any that are good enough to invest time in. Just saying "the machine can do it" isn't enough.
As far as "privacy" / "filtering" features ... once again, no ideas worth the effort/complexity.
----
💡 the solution to where do I write my grocery list is not Atacama. it probably never will be.
When looking at historical chess matches, it is interesting to note that neither player would have had (or understood) computer analysis of positions.
----
It is like the "cell phone" problem. We have a media ⚙️ "Media" includes Shakespeare and Dante, the Arabian Nights and the Red Mansions. but, also, Andy Griffith and John Wayne. based around not being available to everyone 24/7 by cell phone.
----
it is like a religion of old, to have access to Stockfish for chess position analysis. A supernatural knowledge of the situation.
https://www.chess.com/news/view/being-a-chess-professional-in-europe-sucks-blohberger
While players inside the top 20 or 30 "don't really have that many financial struggles," Blohberger said that grandmasters outside of that do, especially in Europe. The two main problems, as he described, were (1) the high cost of living and (b) the lack of recognition.
In both respects, he compared the conditions of Europeans and Asians. To illustrate one difference, he gave an example: imagine two 2600-rated players, one from Germany and the other from India. "If both of these players have the same rating... and they play the same tournaments... they will, on average, make the same prize money, let's say, and this prize money is of course worth way more in this case in India than in Germany."
He also raised that the prizes and "honorary" (a stipend paid by leagues or clubs) have stayed the same for decades and have not risen with inflation.
The second issue was recognition. While chess has been historically popular in Europe, the popularity boom seems to have transferred to Asia.
The short answer is YES. The iron law of economics is that, in a field such as this 💡 revenue is generated from "sponsorship", minimal barriers to entry, legible skill differences, the "marginal" professional will only earn subsistence wages. And this guy, at about 500th in the world (2530 ELO), is clearly marginal.
yesterday:
----
today:
----
other news:
the new "lexer"-based parser has crossed the hump of possibility.
it took roughly 2 days of effort (plus 3 days of complaining, earlier). this is, arguably, slower than it would have been for me to write this without using the machine.
there are a few minor things to fix before it replaces the new system. those are in the next list.
----
today's things:
things done on Atacama yesterday:
----
things to-do today:
other than "get Claude to write a regression test", my attention was on projects other than Atacama yesterday.
----
today, considering 3 features:
things like "SMTP API" simply aren't worth the complexity, much less my time to get the machine to implement it. ⚙️ the compute cost of having the machine implement the SMTP API, presumably, is at most $1-2; it is not a relevant factor. 💡 the price to me is already included in my monthly subscription. the "cost" is different than the "price"
as far as the "quotes" system: it seems like something to leave alone. either let it rot, or wait for the improvements to become more evident.
as far as the LLM features ... 💬 As one grows older, one sees the impossibility of imposing your will on the chaos with brute force. But, if you’re patient, there may come the moment when, while you are eating an apple, the solution presents itself and says, "Here I am."