azergante

• azerganteMar 9, 2025, 10:48 PM

  1 point

  0

  on: How to Corner Liars: A Mi­asma-Clear­ing Protocol

  Planecrash (from Eliezer and Lintamande) seems highly relevant here: the hero, Keltam, tries to determine whether he is in a conspiracy or not. To do that he basically applies Bayes theorem to each new fact he encounters: “Is fact F more likely to happen if I am in a conspiracy or if I am not? hmm, fact F seems more likely to happen if I am not in a conspiracy, let’s update my prior a bit towards the ‘not in a conspiracy’ side”.

  Planecrash is a great walkthrough on how to apply that kind of thinking to evaluate whether someone is bullshitting you or not, by keeping two alternative worlds that explain what they are saying, and updating the likelihoods as the discussion goes on.

  Surely if you start putting probability on events such as “someone stole my phone”, and “that person then tailed me”, and multiply the probability of each new fact, it gets really unlikely really fast. Also relevant: Burdensome details

• azerganteMar 9, 2025, 10:13 PM

  1 point

  0

  on: Lots of brief thoughts on Soft­ware Engineering

  You did not explicitly state the goal of the advice, I think it would be interesting to distinguish between advice that is meant to increase your value to the company, and advice meant to increase your satisfaction with your work, especially when the two point in opposite directions.

  For example it could be that “swallow[ing] your pride and us[ing] that garbage language you hate so much” is good for the company in some cases, but terrible for job satisfaction, making you depressed or angry every time you have to use that silly language/​tool.

• azerganteMar 9, 2025, 10:03 PM

  10 points

  0

  on: Lots of brief thoughts on Soft­ware Engineering

  For that reason try to structure teams such that every team has everything it needs for its day to day work.

  I would extend that to “have as much control as you can over what you do”. I increasingly find that this is key to move fast and produce quality software.

  This applies to code and means dependencies should be owned and open to modifications, so the team understands them well and can fix bugs or add features as needed.

  This avoids ridiculous situations where bugs are never fixed or shipping very simple features (such as changing a theme for a UI component) is impossible or takes weeks because a framework actively prevents it.

  More control and understanding also tends to be better for satisfaction. Of course all this is on a spectrum and should be balanced with other requirements.

• azerganteMar 7, 2025, 7:31 PM

  1 point

  0

  on: az­er­gante’s Shortform

  Tip: you can ask ChatGPT to include confidence scores in its replies

  Interactions with ChatGPT can be customized durably in the options, for example you can add the following instructions: “include a confidence rating at the end of your response in the format ‘Confidence: X%’. If your confidence is below 80%, briefly explain why”.

  Here is a sample conversation demonstrating this and showing what ChatGPT has to say about its calibration:

  Me: Are you calibrated, by which I mean, when you output a confidence X as a percentage, are you right X times out of 100?

  ChatGPT: Reasoned about confidence calibration for 27 seconds. I don’t have calibration in that strict sense. When I mention a confidence percentage, it’s a heuristic internal signal rather than a guarantee that I’ll be right exactly that proportion of the time. Research shows that language models often exhibit miscalibration, meaning a stated confidence (say 70%) doesn’t empirically translate to being correct 70% of the time. I’m about 90% confident in this explanation.

  Me: How bad is the miscalibration?

  ChatGPT: Thought for a second. LLMs—especially those fine‐tuned with RLHF—are often notably overconfident. For example, a model might claim 90% confidence while its actual accuracy is closer to 70–80%. Studies report expected calibration errors (ECE) of roughly 5–10% or more, and effective calibration methods can sometimes cut these errors by about half, underscoring that miscalibration is a significant issue. I’m about 90% confident in this summary, based on multiple recent calibration studies.

  Me: If you are predictably overconfident, and you know it, do you think you can notice that before stating your confidence and correct for it?

  ChatGPT: Reasoned about overconfidence and calibration adjustment for 4 seconds. I don’t have a real-time “internal monitor” to detect my own overconfidence before stating it. Instead, I rely on calibration techniques—like temperature scaling and post-hoc adjustments—that help align my confidence levels with actual accuracy. In other words, I can be tuned to be less overconfident, but I don’t consciously correct my confidence on the fly. I’m about 90% confident in this explanation based on current research findings.

• azerganteMar 5, 2025, 9:48 PM

  11 points

  4

  on: How Much Are LLMs Ac­tu­ally Boost­ing Real-World Pro­gram­mer Pro­duc­tivity?

  Many developers have been reporting that this is dramatically increasing their productivity, up to 5x’ing/​10x’ing it

  I challenge the data: none of my colleagues have been reporting this high a speed-up. I think your observation can just be explained by a high sampling bias.

  People who do not use AI or got no improvement are unlikely to report. You also mention Twitter where users share “hot takes” etc to increase engagement.

  It’s good to have actual numbers before we explain them, so I ran a quick search and found 3 articles that look promising (I only did a basic check on the methodology, don’t take these numbers at face value without analyzing the source in depth):

  • An Axify analysis of the DORA metrics they collect: https://​​axify.io/​​blog/​​use-ai-for-developer-productivity

  Documentation quality (+7.5%)

  Code review speed (+3.1%)

  Delivery throughput (-1.5%): AI adoption slightly decreases delivery throughput, usually due to over-reliance, learning curve, and increased complexity.

  Delivery stability (-7.2%): It is significantly impacted because AI tools can generate incorrect or incomplete code, increasing the risk of production errors.

  What are the DORA metrics?

  Deployment frequency | How often a team puts an item into production.

  Lead time for changes | Time required for a commit to go into production.

  Change failure rate | Percentage of deployments resulting in production failure.

  Failed deployment recovery time | Time required for a team to recover from a production failure.

  • A McKinsey pilot study on 40 of their developers: https://​​www.mckinsey.com/​​capabilities/​​mckinsey-digital/​​our-insights/​​unleashing-developer-productivity-with-generative-ai

  

  • An UpLevel analysis of the data of 800 developers: https://​​uplevelteam.com/​​blog/​​ai-for-developer-productivity

  Analyzing actual engineering data from a sample of nearly 800 developers and objective metrics, such as cycle time, PR throughput, bug rate, and extended working hours (“Always On” time), we found that Copilot access provided no significant change in efficiency metrics.

  The group using Copilot introduced 41% more bugs

  Copilot access didn’t mitigate the risk of burnout

  “The adoption rate is significantly below 100% in all three experiments,” the researchers wrote. “With around 30-40% of the engineers not even trying the product.

  In my experience LLMs are a replacement for search engines (these days, search engines are only good to find info when you already know on which website to look …). They don’t do well in moderately sized code bases, nor in code bases with lots of esoteric business logic, which is to say they don’t do well in most enterprise software.

  I mostly use them as:

  • a documentation that talks: it reminds me of syntax, finds functions in APIs and generates code snippets I can play with to understand APIs

  • a fuzzy version of an encyclopedia: it helps me get an overview of a subject and points me to resources where I can get crisper knowledge

  • support/​better search engine for weird build issues, or weird bugs etc

  I think it’s also good at one shot scripts, such as data wrangling and data viz, but it does not come up often in my current position.

  I would rate the productivity increase at about 10%. I think the use of modal editors (Vim or modern alternatives) improve coding speed more than inline AI completion, which is often distracting.

  A lot of my time is spent understanding what I need to code in a back and forth with the product manager or clients. Then I can either code it myself (it’s not so hard once I understand the requirements) or spend time explaining the AI what it needs to do, watch it write sloppy code, and rewrite the thing.

  Once in a while the coding part is actually the hard part, for example when I need to make the code fast or make a feature play well with the existing architecture, but then the AI can’t do logic well enough to optimize code, nor can it reason about the entire code base which it can’t see anyway.

• azerganteJan 16, 2025, 8:53 PM

  1 point

  0

  in reply to: AnthonyC’s comment on: We prob­a­bly won’t just play sta­tus games with each other af­ter AGI

  I also think it is unlikely that AGIs will compete in human status games. Status games are not just about being the best: Deep Blue is not high status, sportsmen that take drugs to improve their performance are not high status.

  Status games have rules and you only win if you do something impressive while competing within the rules, being an AGI is likely to be seen as an unfair advantage, and thus AIs will be banned from human status games, in the same way that current sports competitions are split by gender and weight.

  Even if they are not banned given their abilities it will be expected that they do much better than humans, it will just be a normal thing, not a high status, impressive thing.

• azerganteJan 6, 2025, 10:02 PM

  1 point

  0

  on: az­er­gante’s Shortform

  Let me show you the ropes

  There is a rope.
  You hold one end.
  I hold the other.
  The rope is tight.
  I pull on it.

  How long until your end of the rope moves?

  What matters is not how long until your end of the rope moves.
  It’s having fun sciencing it!

• azerganteDec 30, 2024, 3:40 PM

  11 points

  0

  on: Learn to write well BEFORE you have some­thing worth saying

  For those interested in writing better trip reports there is a “Guide to Writing Rigorous Reports of Exotic States of Consciousness” at https://​​qri.org/​​blog/​​rigorous-reports

  A trip report is an especially hard case of something one can write about:

  • english does not have a well-developed vocabulary for exotic states of consciousness

  • even if we made up new words, they might not make much sense to people that have not experienced what they point at, just like it’s hard to describe color to blind people or to project a high-dimensional thing to a lower dimensional space.

• azerganteDec 26, 2024, 10:31 PM

  0 points

  0

  in reply to: Roger Scott’s comment on: Biolog­i­cal risk from the mir­ror world

  I have a similar intuition that if mirror-life is dangerous to Earth-life, then the mirror version of mirror-life (that is, Earth-life) should be about equally as dangerous to mirror-life as mirror-life is to Earth-life. Having only read this post and in the absence of any evidence either way this default intuition seems reasonable.

  I find the post alarming and I really wish it had some numbers instead of words like “might” to back up the claims of threat. At the moment my uneducated mental model is that for mirror-life to be a danger it has to:

  • find enough food that fit its chirality to survive

  • not get killed by other life-forms

  • be able to survive Earth temperature, atmosphere etc etc

  • enter our body

  • bypass our immune system

  • be a danger to us

  Hmm, 6 ifs seems like a lot, so is it unlikely? in the absence of any odds it is hard to say.

  The post would be more convincing and useful if it included a more detailed threat model, or some probabilities, or a simulation, or anything quantified.

  A last question: how many mirror molecules does an organism need to be mirror-life? is one enough? does it make any difference to its threat-level?

• azerganteDec 23, 2024, 12:54 PM

  5 points

  2

  on: az­er­gante’s Shortform

  2+2=5 is Fine Maths: all you need is Coherence

  [ epistemological status: a thought I had while reading about Russell’s paradox, rewritten and expanded on by Claude ; my math level: undergraduate-ish ]

  Introduction

  Mathematics has faced several apparent “crises” throughout history that seemed to threaten its very foundations. However, these crises largely dissolve when we recognize a simple truth: mathematics consists of coherent systems designed for specific purposes, rather than a single universal “true” mathematics. This perspective shift—from seeing mathematics as the discovery of absolute truth to viewing it as the creation of coherent and sometimes useful logical systems—resolves many historical paradoxes and controversies.

  The Key Insight

  The only fundamental requirement for a mathematical system is internal coherence—it must operate according to consistent rules without contradicting itself. A system need not:

  • Apply to every conceivable case

  • Match physical reality

  • Be the “one true” way to approach a problem

  Just as a carpenter might choose different tools for different jobs, mathematicians can work with different systems depending on their needs. This insight resolves numerous historical “crises” in mathematics.

  Historical Examples

  The Non-Euclidean Revelation

  For two millennia, mathematicians struggled to prove Euclid’s parallel postulate from his other axioms. The discovery that you could create perfectly consistent geometries where parallel lines behave differently initially seemed to threaten the foundations of geometry itself. How could there be multiple “true” geometries? The resolution? Different geometric systems serve different purposes:

  • Euclidean geometry works perfectly for everyday human-scale calculations

  • Spherical geometry proves invaluable for navigation on planetary surfaces

  • Hyperbolic geometry finds applications in relativity theory

  None of these systems is “more true” than the others—they’re different tools for different jobs.

  Russell’s Paradox and Set Theory

  Consider the set of all sets that don’t contain themselves. Does this set contain itself? If it does, it shouldn’t; if it doesn’t, it should. This paradox seemed to threaten the foundations of set theory and logic itself.

  The solution was elegantly simple: we don’t need a set theory that can handle every conceivable set definition. Modern set theories (like ZFC) simply exclude problematic cases while remaining perfectly useful for mathematics. This isn’t a weakness—it’s a feature. A hammer doesn’t need to be able to tighten screws to be an excellent hammer.

  The Calculus Controversy

  Early calculus used “infinitesimals”—infinitely small quantities—in ways that seemed logically questionable. Rather than this destroying calculus, mathematics evolved multiple rigorous frameworks:

  • Standard analysis using limits

  • Non-standard analysis with hyperreal numbers

  • Smooth infinitesimal analysis

  Each approach has its advantages for different applications, and all are internally coherent.

  Implications for Modern Mathematics

  This perspective—that mathematics consists of various coherent systems with different domains of applicability—aligns perfectly with modern mathematical practice. Mathematicians routinely work with different systems depending on their needs:

  • A number theorist might work with different number systems

  • A geometer might switch between different geometric frameworks

  • A logician might use different logical systems

  None of these choices imply that other options are “wrong”—just that they’re less useful for the particular problem at hand.

  The Parallel with Physics

  This view of mathematics parallels modern physics, where seemingly incompatible theories (quantum mechanics and general relativity) can coexist because each is useful in its domain. We don’t need a “theory of everything” to do useful physics, and we don’t need a universal mathematics to do useful mathematics.

  Conclusion

  The recurring “crises” in mathematical foundations largely stem from an overly rigid view of what mathematics should be. By recognizing mathematics as a collection of coherent tools rather than a search for absolute truth, these crises dissolve into mere stepping stones in our understanding of mathematical systems.

  Mathematics isn’t about discovering the one true system—it’s about creating useful systems that help us understand and manipulate abstract patterns. The only real requirement is internal coherence, and the main criterion for choosing between systems is their utility for the task at hand.

  This perspective not only resolves historical controversies but also liberates us to create and explore new mathematical systems without worrying about whether they’re “really true.” The question isn’t truth—it’s coherence.

az­er­gante’s Shortform

• azerganteNov 23, 2024, 4:11 PM

  1 point

  0

  on: Ideas for an ac­tion co­or­di­na­tion website

  I really like the idea of milestones, I think seeing the result of each milestones will help create trust in the group, confidence that the end action will succeed and a realization of the real impact the group has. Each CA should probably start with small milestones (posting something on social medias) and ramp things up until the end goal is reached. Seeing actual impact early will definitely keep people engaged and might make the group more cohesive and ambitious.

• azerganteNov 23, 2024, 3:50 PM

  1 point

  0

  on: If a “Kick­starter for Inad­e­quate Equlibria” was built, do you have a con­crete in­ad­e­quate equil­ibrium to fix?

  Ditch old software tools or programming languages for better, new ones.

[Question] A Co­or­di­na­tion Cook­book?

• azerganteAug 30, 2024, 7:26 PM

  1 point

  0

  on: Su­per­in­tel­li­gence 16: Tool AIs

  My take on the tool VS agent distinction:

  • A tool runs a predefined algorithm whose outputs are in a narrow, well-understood and obviously safe space.

  • An agent runs an algorithm that allows it to compose and execute its own algorithm (choose actions) to maximize its utility function (get closer to its goal). If the agent can compose enough actions from a large enough set, the output of the new algorithm is wildly unpredictable and potentially catastrophic.

  This hints that we can build safe agents by carefully curating the set of actions it chooses from so that any algorithm composed from the set produces an output that is in a safe space.

• azerganteMay 12, 2024, 9:08 PM

  4 points

  0

  on: Deep Honesty

  I think being as honest as reasonably sensible is good for oneself. Being honest applies pressure on oneself and one’s environment until the both closely match. I expect the process to have its ups and downs but to lead to a smoother life on the long run.

  An example that comes to mind is the necessity to open up to have meaningful relationships (versus the alternative of concealing one’s interests which tends to make conversations boring).

  Also honesty seems like a requirement to have an accurate map of reality: having snappy and accurate feedback is essential to good learning, but if one lies and distorts reality to accomplish one’s goals, reality will send back distorted feedback causing incorrect updates of one’s beliefs.

  On another note: this post immediately reminded me of the buddhist concept of Right Speech, which might be worth investigating for further advice on how to practice this. A few quotes:

  “Right speech, explained in negative terms, means avoiding four types of harmful speech: lies (words spoken with the intent of misrepresenting the truth); divisive speech (spoken with the intent of creating rifts between people); harsh speech (spoken with the intent of hurting another person’s feelings); and idle chatter (spoken with no purposeful intent at all).”

  “In positive terms, right speech means speaking in ways that are trustworthy, harmonious, comforting, and worth taking to heart. When you make a practice of these positive forms of right speech, your words become a gift to others. In response, other people will start listening more to what you say, and will be more likely to respond in kind. This gives you a sense of the power of your actions: the way you act in the present moment does shape the world of your experience.”

  Thanissaro Bhikkhu (source: https://​​www.accesstoinsight.org/​​lib/​​authors/​​thanissaro/​​speech.html)

• azerganteMay 9, 2024, 9:39 PM

  1 point

  0

  in reply to: vizikahn’s comment on: Talk­ing Snakes: A Cau­tion­ary Tale

  I also thought about something along those lines: explaining the domestication of wolves to dogs, or maybe prehistoric wheat to modern wheat, then extrapolating to chimps. Then I had a dangerous thought, what would happen if we tried to select chimps for humaneness?

• azerganteMay 8, 2024, 6:40 PM

  1 point

  0

  on: Se­crets of the eliminati

  goals appear only when you make rough generalizations from its behavior in limited cases.

  I am surprised no one brought up the usual map /​ territory distinction. In this case the territory is the set of observed behaviors. Humans look at the territory and with their limited processing power they produce a compressed and lossy map, here called the goal.

  The goal is a useful model to talk simply about the set of behaviors, but has no existence outside the head of people discussing it.

• azerganteMay 7, 2024, 9:06 PM

  1 point

  0

  in reply to: papetoast’s comment on: Ask­ing for help as an O(1) lookup

  This is a great use case for AI: expert knowledge tailored precisely to one’s needs

• azerganteApr 27, 2024, 6:34 PM

  1 point

  −1

  on: Su­per­in­tel­li­gence FAQ

  Is the “cure cancer goal ends up as a nuke humanity action” hypothesis valid and backed by evidence?

  My understanding is that the meaning of the “cure cancer” sentence can be represented as a point in a high-dimensional meaning space, which I expect to be pretty far from the “nuke humanity” point.

  For example “cure cancer” would be highly associated with saving lots of lives and positive sentiments, while “nuke humanity” would have the exact opposite associations, positioning it far away from “cure cancer”.

  A good design might specify that if the two goals are sufficiently far away they are not interchangeable. This could be modeled in the AI as an exponential decrease of the reward based on the distance between the meaning of the goal and the meaning of the action.

  Does this make any sense? (I have a feeling I might be mixing concepts coming from different types of AI)