Saved on Hacker News · submitted by reasonableklout
Life on Claude Nine (babuschk.in)
Drones built to look like birds evaded detection at Air Force tech exercise (taskandpurpose.com)
Posts by other people that I saved on Twitter and Hacker News, by the date they were posted. Each one links to the original.
5632 bookmarks · page 1 of 226
Life on Claude Nine (babuschk.in)
Drones built to look like birds evaded detection at Air Force tech exercise (taskandpurpose.com)
Verification Is Not Knowledge: Science needs more than answers (blog.cosmos-institute.org)
Asked Claude to replicate this and it works 😬 though you do need contiguous audio and how much you need is very location dependent (🧵)
Set Theory Symbols.
Some Important Notations in Calculus.
Ja, es sind gefährliche Leute, mit denen sich Teile der politischen Linken in Berlin eingelassen haben. Und wir haben Verständnis dafür, dass Elif Eralp und ihr Mann »Angst vor diesen Leuten« haben, wie die Linken-Kandidatin für das Amt der Regierenden Bürgermeisterin in einem intern kommentierten Interviewdokument schreibt, das wohl versehentlich an @derspiegel gesendet wurde. Gemeint ist laut SPIEGEL das Milieu um Ibrahim Ibrahim und das Vereinigte Palästinensische Nationalkomitee (VPNK), das der Berliner Verfassungsschutz als extremistisch einstuft und als Dachorganisation für Anhänger von PFLP und Hamas beschreibt.
Eralp strich aus der Autorisierung die SPIEGEL-Frage »Damit meinen Sie den Terrorsympathisanten Ibrahim Ibrahim, der den Chefplaner des Hamas-Überfalls auf Israel am 7. Oktober 2023 als ›Märtyrerhelden‹ bezeichnet hat und in Neukölln auf der Wahlparty war.« Gestrichen wurde auch ihr eigenes »Ja«.
In der Randnotiz schreibt Eralp: »Ich hätte dieses Interview niemals zusagen dürfen.« Der SPIEGEL weiter: »Dann nimmt sie Bezug auf eine Empfehlung, die ihr offenbar das Landeskriminalamt in der Vergangenheit gegeben hat, als es um ihren öffentlichen Umgang mit dem Thema Organisierte Kriminalität (OK) ging. ›Das LKA hat damals gesagt, sie empfehlen, nie konkrete Namen zu nennen. Damals ging es um OK. Jetzt geht’s um Leute mit Kontakten zur Hamas. Ich kann mich und meine Familie für den SPIEGEL nicht in Gefahr bringen. Was machen wir jetzt? Das ganze Interview stornieren?‹«
Zugleich zeigt der SPIEGEL, dass Teile ihrer Partei die Nähe zu Ibrahim suchen: Die LAG Palästinasolidarität der Berliner Linken erklärte »unserem Genossen Ibrahim Ibrahim« ihre volle Solidarität, der Bezirksverband Neukölln sitzt mit dem VPNK im selben Bündnis, Ibrahim posierte auf der Linken-Wahlparty mit Ferat Koçak und Vanessa Emde.
Der SPIEGEL stellt die Fragen, die jetzt über den Sondierungen stehen: Kann eine Frau, die zu viel Angst hat, um den Namen eines Terrorsympathisanten zu bestätigen, für die Sicherheit aller Berlinerinnen und Berliner verantwortlich sein? Und wer würde am Ende wen regieren?
Zum Bericht von Christoph Hickmann, Marina Kormbaki, Ulrich Kraetzer, Anna Reimann, Marc Röhlig und Hannes Schrader:
"AI skills" are the same skills we always needed:
- Learning in a self-directed way that aligns with the needs of your role and where it is going
- Clear communication, the ability to edit and know how to say what is important
- Forming and testing hypotheses
- Evaluating and testing tools and work product, and being able to recognize and rectify deviations
- Delegating, setting constraints, rubrics and escalation criteria
- Having a clear definition of done and being able to stick with it
- Understanding business domains and modelling them usefully
- Collaborating with others, communicating your needs and capabilities clearly, eliciting the needs and capabilities of others, synthesizing conflicting needs and aligning compounding capabilities
The adoption of genAI and agentic tooling really just serves to make all of these more important.
He advocated an immediate preemptive nuclear strike on the USSR: "If you say why not bomb them tomorrow, I say why not today? If you say today at five o'clock, I say why not one o'clock?"
This is why he was dismissed in so far as foreign policy was concerned, where nuance and patience takes precedence over cold, calculating game theory. Not only would it not have been feasible (Russia is huge), but it would have led to follow on effects by possibly pulling the US into potentially another world war. This is evidence you can be a genius in some areas and wrong in others.
@thegreatest_sv Cool idea on paper, but your recent posts sit well below 100k impressions. Experimental tools are fine- just stop selling them like guaranteed viral machines when the numbers and repo say otherwise.
Made a skill for dimensional drawings of UI components.
Inspired by Apple's accessory drawings, it documents a component's architecture into a spec sheet.
You can make serious progress training almost anything in 30 high-intensity minutes per day.
For instance: If you worked at a pace of 30 XP/day, every day, then you could complete our entire Mathematical Foundations sequence -- from fractions to calculus -- all in a bit over a year.
(There are 3 courses, each ~5000 XP, roughly ~15000 XP or 15000 minutes in total. We benchmark 1 XP ~ 1 minute of fully focused work for an average serious but imperfect student; any individual student may move a bit slower or faster depending on their performance.)
My bar for AGI is still an AI winning a Nobel.
But an AI researcher looking at the lm_head (that everyone tells you to keep in high precision) deciding to quantize it to 4 bits AND being actually right (0.7pt from bf16) is definitely impressive.
Even better when the result are open-weights pushed on the Hugging Gace hub.
Congrats @Recursive_SI !
Dear god it's inventing German from first principles
If you have to pitch it’s already over
Google just open-sourced Harness².
(turn any harness into a self-improving system)
a harness is everything wrapped around the model, like the system prompt, tools, memory, skills, and subagents. today a person designs it once, and the agent runs every new task with that same fixed setup.
Harness² lets the agent refine that setup for each new task on its own, with the model weights left untouched.
on every task, the agent proposes edits to its harness, runs a few candidate versions, and composes a better harness by comparing how those runs behaved.
because it learns from the differences between runs, it needs no correct answer or grader for the task.
that covers a single task. Harness² also carries what it learns from one task into the next.
after each task, it saves the harness it settled on along with the lessons that held up across the runs. the next time it refines a harness, it starts from those notes instead of from scratch.
this means every finished task makes the next round of edits more informed.
Google tested this with Gemini 3.5 Flash and GLM-5 on legal, office, and professional work benchmarks. one round of edits was enough to beat their original harnesses, and giving the agent more rounds pushed scores higher.
those extra rounds can go wide or deep. the agent either tries several candidate harnesses at once or keeps refining a single one step by step, using OpenCode or Codex to carry out the work.
you can find the Harness² repo here: github.com/google-researc…
I wrote about a similar idea for agents already running in production, where the harness should repair itself after a failure instead of waiting for someone to debug the trace by hand.
that piece walks through how @Cometml's Opik (open-source) closes this loop, going from a failing trace to a proposed fix to a regression test that stops the same failure from coming back. it's quoted below.
My dear friends, I’d like to show you my latest toy project
Introducing Handwired: turn your regular JavaScript code into a neural net!
All runnable in a browser, without GPU
I'm a bit shocked that our RSI system (AIDE²) actually discovered a similar mechanism.
I initially thought it was a weird mechanism that had slipped through due to some lucky evaluation noise. But after reading this more rigorous study, it now makes sense to me.
Friendly reminder: the world is run by high agency people, not smart people. Stop consuming information and start taking insane levels of action.
Andrej Karpathy spent 8 years at OpenAI and Tesla
Last week he compressed everything he knows into one free 2-hour lecture
Prompts → Agents → Harness → Graphs → Loops
People spend $15k on bootcamps that teach less than this
You probably don't have 2 hours right now
Don't let it vanish from your feed
Watch it, then read the full Opus 5.5 harness engineering guide below
The idea of an intelligence explosion caused by recursive self improvement has been around for a long time but until very recently it did not seem imminent. Now many leading researchers think it may happen quite soon. You can read our paper about it here:
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
what doesn't kill you introduces you to carl jung, marcus aurelius, epictetus, friedrich nietzsche, james baldwin, bell hooks, audre lorde, frantz fanon, paulo freire, joseph campbell, alan watts, ram dass, gary zukav, brene brown, gabor mate, shadow work, attachment theory, nervous-system regulation, ancestral trauma, existentialism, stoicism, meditation, emotional intelligence, neuroplasticity, metacognition, pattern recognition, game theory, delusional optimism, boundaries and discernment.
If you're using AI all day and you're not making more money, it means you have the same problem you had before the AI: you're doing stuff that doesn't matter.
This is the real reason 99% of people don't get ahead. They don't know how to tell the difference between what's important and what's irrelevant. And just doing irrelevant things faster gets you the same outcome: nothing.
Except this time it cost you tokens.