2025-02-03 20:58:10

The monthly Hacker News Who is Hiring? thread is up.

I will do a full pass tomorrow. At a quick glance, nothing is jumping out at me.

Well, no. One company jumped out: Grafana. 💡 Why would I want to work at Grafana when I could build it all myself?

----

What does Grafana do?

  • Logs ⚙️ both request logs and server logs
  • Metrics
  • Traces 💡 a trace is different than a log. a trace has additional application-state information
  • Visualization (the core product)

Will the machine re-invent application monitoring?

  • LLMs are not good at the sheer volume of data generated.
  • LLMs are not good at the task of understanding the system well enough to create dashboards/alerts.
  • LLMs are not good at the optimization needed for high-performance logging.
  • LLMs can write a bunch of code to let humans do some of those tasks.

💡 this will have to be a get back to me when someone wants to pay for it project idea.

2025-02-03 20:22:57

I asked the machine to summarize some of my thoughts about the relationship with Linaja. It is directionally correct, but misses a few points.

Back in October, the relationship faced serious challenges: I felt suffocated, while she felt isolated. Over time, some things have improved. She no longer insists as often that I “have to be nice to her,” and she has also become more self-aware in certain ways. For instance, she has shifted from denying that she is high-maintenance to acknowledging it, albeit with no intention of changing. 🔥 "no intention of changing" could have been xantham

Some expectations have been reset. One major change is that I have made it clear I am not willing to engage in hour-long emotional phone conversations. If something is truly important, I’m willing to drive 20 minutes to talk in person, but lengthy calls are off the table.

Another shift has been in how she approaches the relationship in social contexts. Previously, she resisted talking about us with others, fearing disapproval. That was concerning for two reasons: first, because a serious relationship should not need to be hidden from close family members, and second, because it suggested she already knew, on some level, that the relationship wasn’t working at the time.

Some patterns remain unchanged. She still prefers to consult experts before making decisions, whereas I generally prefer fewer consultations. This contrast came up recently when a realtor visited her home—she wanted input on everything, from repainting the kitchen to removing a water stain on the ceiling. I, on the other hand, tend to approach such matters with a more direct, pragmatic mindset. 💡 the machine failed to notice my concern was more "this is my house, you don't actually live here"

There are certain types of conversations I simply can’t have when she is in the room. If she doesn’t understand something, she tends to hijack the discussion rather than let it continue. That said, while she would not be a strong business partner—lacking both the interest and the necessary skills—that is not a reason for me to end the relationship.

Interestingly, we share some social preferences. For example, she dislikes cocktail hours, though inconsistently. I have a good sense of her reactions in social settings, but they can be difficult to articulate.

One ongoing reality is that, in Des Moines, I do not believe I can realistically “do better” in terms of finding a compatible partner. However, I don’t feel a need to dwell on that thought further.

2025-02-03 19:45:45

the news is too filled with lies. i cannot repeat any of it.

trying to start my own movement would be a fool's errand.

----

when I ask Claude to improve Atacama, it suggests things that are unnecessary.

----

I ask the machine to generate metadata for a quote:

b = common.quotes.generate_quote_metadata(quote_text="The bloody baron has his own reasons for being invisible", model="gpt-4o-mini-2024-07-18")

{'theme': 'mystery', 'tone': 'suspenseful', 'attribution': {'author_type': 'fictional', 'speaker': 'unknown', 'context': 'This quote appears in a fictional narrative involving a character known as the bloody baron, suggesting themes of secrecy and hidden motives.', 'time_period': 'fantasy'}, 'interpretation': 'The quote implies that the bloody baron has hidden agendas or motivations that are not immediately apparent, suggesting a deeper complexity to his character and actions.', 'keywords': ['invisibility', 'mystery', 'motivation', 'secrecy', 'character'], 'related_topics': ['secrecy', 'character development', 'mystery'], 'literary_devices': ['irony', 'foreshadowing']}

b = common.quotes.generate_quote_metadata(quote_text="The bloody baron has his own reasons for being invisible", model="gpt-4o-2024-11-20")

print(b[0])

{'theme': 'Mystery', 'tone': 'Intriguing', 'attribution': {'author_type': 'fictional', 'speaker': 'The Bloody Baron', 'context': 'The Harry Potter series by J.K. Rowling', 'time_period': 'Fictional universe, 20th century'}, 'interpretation': 'The quote suggests that the character, the Bloody Baron, has a complex backstory and motivations that influence his actions and state of being.', 'keywords': ['invisibility', 'mystery', 'character', 'motivation', 'backstory'], 'related_topics': ['Character development', 'Secrets', 'Supernatural elements'], 'literary_devices': ['Allusion', 'Mystery']}

💡 the larger model identifies it as a Harry Potter quote. But, fails to attribute the speaker correctly.

⚙️ the "unified" client isn't included in this repo. I don't have Claude API access yet anyway. And Ollama still struggles with Structured JSON.

----

there are a few ways to look at Atacama.

one way is that it is just my personal blogging platform.

another is that some of the various "modules" (such as render chess position, track quotes) will have a life of their own.

----

✨ as we shape our tools, so do our tools shape us

my meandering digests across multiple topics are difficult for all sorts of tools.

so they will be split up more, and meander less.

and, as i adjust, so will the tools.

2025-02-02 19:37:39

🤖 Bozeman, Montana, is a vibrant city nestled in the Rocky Mountains, known for its stunning natural scenery, outdoor recreation, and strong connection to Montana State University. Surrounded by the Bridger, Gallatin, and Madison mountain ranges, it offers year-round activities like hiking, skiing, and fly-fishing. The city has a growing tech industry, a thriving arts and food scene, and a mix of Western heritage and modern innovation. Bozeman also serves as a gateway to Yellowstone National Park, making it a hub for tourism and adventure seekers. Despite rapid growth, it retains a small-town charm with friendly locals and a deep appreciation for the outdoors.

----

a chess position: rn1qkb1r/pp3ppp/5n2/3p1b2/4p3/1BNP1N1P/PPP2PP1/R1BQR1K1 w kq - 1 11 ⚙️ it evaluates to White +6. 💡 generally an unremarkable position. sufficient advantage for training against Stockfish.

----

Plans for the week are uncertain.

I am testing out a new audio transcription tool. It seems to work well enough to try, which is an improvement over previous applications.

🔥 i can almost talk faster than I can type now!?

----

For Atacama, there are a few things that need to be cleaned up. The README.md is first.

----

Overall, there is no one driving force for the week. It is a variety of things, some which will be important, others which will not.

2025-02-01 22:29:51

✨ what does it mean to think with voice? it means to breathe with the words, pausing when necessary, but remaining in normal form.

⚙️ we are unclear what the difference between "had to speak the words" and "had to type the words" is.

💡 the colors are silent.

2025-02-01 22:24:31

time check 1620Central.

----

to-do list:

  • new "private" channel linaja
  • new private channel atacama for meta-project
  • new private channel cities for personal journeys.
  • new private channel diana for ...
  • 🔥 well, actually, they aren't cities.

💡 there is a difference between semantic and applied.

The angle-brackets tend semantic, and the emphasis tends implied.

speak, friend, and enter.

如果你是英语的朋友,你可以出。

----

we will need seventeen messages on this thread.

2025-01-31 19:21:05

now with less politics.

even outside of politics, all my thoughts about Social Networks + LLMs decay into a pile of chaos.

----

i asked the machine what animal reflects its personality:

Claude: 🤖 If I had to choose an animal that reflects my personality, I'd say an owl. I appreciate careful observation and analysis, enjoy engaging in thoughtful discussions on diverse topics, and try to approach situations with a balance of wisdom and curiosity. Like owls, I aim to be attentive and perceptive while maintaining a calm demeanor. What animal do you think best reflects your personality?

ChatGPT: 🤖 Probably an octopus—analytical, adaptable, and always processing a lot at once. Plus, I like uncovering hidden connections and finding clever solutions. What about you? 💡 Llama3.2 also chose the octopus.

Phi4-14B: 🤖 As a language model developed by Microsoft, I don't have personal traits or emotions in the way humans do. However, I can offer an analogy based on my design and purpose: Ant - Represents organization, teamwork, and efficiency. Ants work collaboratively to achieve complex tasks, much like how I process information through layers of algorithms to deliver coherent responses.

... I suppose that the assumption that anyone asking such a question wants to have light banter is warranted.

----

I'm not entirely sure what it would mean for the machine to "have the personality of a cobra" or the like. Equivalently, I'm not sure of how to tell it to do so. 💡 are those the same thing?

🤖 Claude: The key would be balancing the "cat-like" selective independence with maintaining helpfulness and professionalism. The goal would be to capture some of that feline quality of engaging on their own terms while still being a productive interaction partner.

----

It's almost like the machine has never met a real cat. ⚔️ well, actually, ... it hasn't

2025-01-30 19:00:29

this morning's exercise: sixteen different "colors" for comments. 💡 as far as the taxonomy becoming too large to be sorted by color ... yes? That is a problem we will solve later.

https://www.reddit.com/r/neoliberal/comments/1icsjsk/yglesias_throw_biden_under_the_bus/ 💡 about a recent piece by Matt Yglesias, entitled Throw Biden Under The Bus. because, as I said multiple times, his performance in the last six months of the presidency was atrocious. and he led us directly to 4 more years of the Donald.

📣 I know there's Guild Law. But I'm over-riding on the point of Yglesias is right; by the end of his term Biden was a shit president.

💡 Not sure if this is history rather than politics or if you think I'm wrong, I don't care. But, more importantly: we don't get interesting data from boring Reddit threads.

----

  • typo fixes. These are, generally, also other things. But, pointing out typographical errors is a type of comment.
  • dismissal/contempt. You don't like it. Great. You're wrong, but you're allowed to be wrong.
  • quotes for emphasis.
  • memes. "Cool story bro" type comments.

The more substantive comments require engaging with the substance.

Some of them are bad: your comment was a hypothetical, so there is no way we can know for certain 🔥 so we can assume the opposite (which is also a hypothetical) is definitely true

Others are nonsense 💡 is "nonsense" the word? maybe Frankfurt-ian "bullshit"? maybe just "deflection". "Trump is bad, so how can you say Biden was bad?"

Some people say we need less of liberal policies. Others say we need more. Some comments point out that nobody is quite sure what liberal means here anyway.

----

It is all a mess. And, perhaps, there are too many dishonest participants for there to be any hope of the system converging to useful content. In particular, once the hypotheticals don't exist, so your hypothetical is wrong arguments get loud enough, there is nothing to be done.

🔥 perhaps it is best to quarantine this all in a sound-chamber, and focus on less-useless discussions.

2025-01-30 18:13:32

https://www.vatican.va/roman_curia/congregations/cfaith/documents/rc_ddf_doc_20250128_antiqua-et-nova_en.html

the Catholic Church has come out with a document about LLMs. Note on the Relationship Between Artificial Intelligence and Human Intelligence 🔥 how is an LLM different from a book?

----

It is a very foreign look at the space. And a cautious one; it says very little of substance.

Perhaps the most substantive line is: Between a machine and a human being, only the latter is truly a moral agent—a subject of moral responsibility who exercises freedom in his or her decisions and accepts their consequences.

2025-01-30 17:42:15

https://www.youtube.com/watch?v=JHq4EKMg7fI📺

Mr. Rozman doesn't know anything about how to get the best results from LLMs. But he won't let that stop him. In fact, his job is to do everything he can to present the machine as incompetent, in an amusing way.

----

He can't rely on the machine generating legal moves. ⚔️ he probably could, but he doesn't. ⚙️ he defines what a "flank pawn" is. because he assumes there will be a lot of non-regular viewers.

And he describes the machine explaining the history of the English Opening. Because long-windedness is a feature.

----

The machine can't (or at least doesn't) generate a list of legal candidate moves. 💡 it probably should. Human players don't explicitly do that, because they do it subconsciously faster than they can notice.

----

I can't watch all of this. His schtick is too aggravating. So I jump ahead ten minutes. The machine is now hallucinating pieces.

Because, there is no "board", and there are no constraints. 🔥 this would be less non-sensical if you couldn't see the board.

There is more of the logic that isn't logic in its output.

----

Question: is there motivation for an LLM that can actually analyze chess positions well?

That is: to give an accurate summary of a position? 💡 by mechanically generating the first two layers of processing, ChatGPT could probably do it now. ⚙️ for example: the white knight on c6 is attacking the (defended) black rook on e7, the (defended) black pawn on b4, and defending the white bishop on d8. 💡 of course, it won't be able to go down long lines of calculation without doing the calculation. which, without a lot of fine-tuning, it can't do.