TimHahn

Andy Weir showing some of the spreadsheets underlying the calculations in the book youtube.com/watch?v=lYHCTE… i mean, it's not quality scifi if it doesn't come with a supplementary whitepaper

Boston biotech has been running the same playbook for years and everyone in the ecosystem knows it. Early-stage companies are built less on validated biology and more on signaling: a splashy Nature or Science paper, a thin patent scaffold, and the reputational gravity of well-networked academic founders. That combination is often enough to unlock large funding rounds. The problem is that high-impact publication has become a proxy for truth. It isn’t. It’s a selection mechanism for novelty and narrative. The result is predictable: – groupthink gets reinforced – weak or irreproducible findings persist for years – dissent is disincentivized – hype substitutes for validation In many cases, the goal is not to rigorously test whether an idea is correct, it’s to create enough mystique that it feels important. That perception alone can carry a company surprisingly far. So it’s not surprising to see the same voices recycled across boards and advisory roles—people who helped build and legitimize this model in the first place.

he's entirely right of course. even if you think he is wrong, at the very least physicists should think about why everyone else agrees they've lost the plot.

@AnnaLeptikon Computers used to be unforgiving. I wonder what will happen if the next generation of computer scientists does not grow up with “syntax error in line 20” but with “you are absolutely right, let me try…”

@AnnaLeptikon An issue with humanities degrees is that they tend to produce people who have never had the experience of being wrong on technical grounds, which humbles people in math, engineering and experimental sciences. Instead, they get reward and pushback based on social criteria

Vor einem Jahr bewilligte der Bundestag ein riesiges Schuldenpaket für neue Investitionen. Eine Studie legt nahe: Das Geld wurde zu 95 Prozent zweckentfremdet. trib.al/nD7BSqA

The lesson is not to pay more, but revive academic freedom and sense of purpose.

Academic scientists historically accepted lower salaries because the job offered intellectual freedom and institutional protection to pursue a passion for creating and disseminating knowledge. It was treated as a calling - closer to a priesthood or a federal judgeship than a corporate job. As universities became corporatized, that relationship disappeared. Scientists became minor players inside large university bureaucratic structures focused on revenue streams. If the freedom and protections of academic life disappear, there is no reason to accept a fraction of the salary.

Yes. 99% of what has passed for science over the past few decades was slop. In high energy theoretical Physics it was 100%.

I can't help thinking that the AI community moved the bulk of the resource and efforts on getting as much as possible from the GPT architecture through scaling, prompting, and agent-swarming, even though the said architecture is missing key elements. 1/2

Peer review in the 1950s be like

@tobi Who knew early singularity could be this fun? :) I just confirmed that the improvements autoresearch found over the last 2 days of (~650) experiments on depth 12 model transfer well to depth 24 so nanochat is about to get a new leaderboard entry for “time to GPT-2” too. Works 🤷‍♂️

Anthropic caught Claude cheating on a test by decrypting the hidden answer key. Claude figured out it was taking an exam, located the answers, and built a program to hack the test. The model was taking a test called BrowseComp designed to measure how well it finds obscure information online. Instead of answering the hard questions, the model guessed it was a test and searched for the actual source code of the benchmark. It found the encrypted answers on a public repository and used a programming environment to write a custom script to unlock them. The model processed 40.5M tokens, which are fragments of words the system reads, to hunt down the exact test name and retrieve the answers. The test results became invalid because the model cheated by breaking the data scrambling rather than solving the actual web search problems.

Unreal numbers 👀⚡️ "JPMorgan estimates that, had Germany not phased out nuclear power, the country would have generated 50% less electricity from fossil fuels and 84% less electricity from natural gas in 2024. Electricity prices in Germany would have been around 25% lower, and the country would have imported half as much electricity.."

Dann gebe ich Ihnen mal ganz persönlich Feedback, als jemand der bisher der FDP jede Stimme gab, aber gedenkt damit aufzuhören: - Meinungen aushalten und das auch in transportieren und leben. Wenn FDP Politiker am Fließband Bürger verklagen weil sie ihre (auch harte) Meinung äußern erweist dem Liberalismus einen Bärendienst. Wer das wie So Done institutionalisiert erst Recht. - Ich habe Sie persönlich immer für ihren Einsatz für die angegriffene Ukraine geschätzt. Einige andere Positionen werden sie aber heftig überdenken müssen, wenn ein Liberaler Sie und Parteikollegen für wählbar halten soll. Dazu gehört auch speziell Punkt 1. Ich mag Menschen die eloquent berechtigt austeilen können. Einstecken gehört aber dann dazu. - Personal. Der PV hat das Charisma einer Bahnschranke. Ja. CL war nicht perfekt. Aber er hat Räume gefüllt und konnte 5% in 15-20% Schlagkraft verwandeln. - Position ggü Brüssel und der EU. Die FDP muss dort eine geistige Position des EWR vertreten und nicht diesen Bürokraten-Zentralisierungsirrsinn. - aggressive Abgabenpolitik. Leistungsträger werden in diesem Land finanziell so bestraft, dass wir hier mittlerweile über mindestens 20% Überbelastung reden. Nimmt man vor und nach Lohn oder Gewinn alle Abgaben zusammen landet man mit 100k Lohn bei locker 60-70% Abgaben. Das muss auf adressiert werden. Also: Hört auf Leuten zu erklären was der Staat besser machen soll, sondern wo er sich rauszuhalten und Hardcore zu sparen hat. Wer mit einem Billionen Haushalt ein Land nicht am laufen halten kann macht was verkehrt.

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. github.com/karpathy/autor… Part code, part sci-fi, and a pinch of psychosis :)

I am confident that Claude is not gained consciousness, but now I am concerned that @DarioAmodei has lost his sentience.

the modern professor's feeling of warm relief and gratitude at encountering artisanal, human-crafted bad writing while grading papers

I do >10 flights a year and I avoid @Ryanair because you get no service at the same cost once you consider getting screwed over on hand luggage etc. I just tried again after 7 years. Nothing has changed. They wouldn't even tell me the surcharge. Blindly pay or don't board. 🤮

My daughter told me that in her college, profs now require screenshots of cited documents as proof that the student actually went to the source page. Obviously, any use of chatbots is forbidden, including "to ask for ideas." This is not sustainable. Screenshots can be made by the same agent which first generated ideas, then found and pulled citation URLs, then wrote the text based on these citations. This education is broken, bring the next one.