Professor of Law & Vice Dean International, King's College London • Global Professor of Law, Notre Dame Law School • Wiener Schnitzel
London / Wien
Joined July 2011
- Tweets11.1K
- Following4.2K
- Followers2.2K
- Likes23.3K
Christoph Kletzer retweeted
Not only have conservatives become vanishingly rare in academia, so have centrists. That’s how complete the left’s dominance is: Even moderates are now a fringe group in academia.
stevestewartwilliams.com/p/a…
Christoph Kletzer retweeted
Unlike many, Florence is an actual anthropologist in the French tradition of Marcel Mauss. Many should read her and pay attention to what she says on the future of Europe.
Christoph Kletzer retweeted
Only 46 measures. Approximately 4 minutes. Mozart wrote “Ave Verum Corpus” in June 1791 — near the end of his life. So simple it's almost bare, yet it contains the heavens. Leonardo Bernstein conducted it at Waldsassen Cathedral in 1990 — slowly, lovingly, as if unwilling to leave each note. A miracle in a small package.
Christoph Kletzer retweeted
Wenn Kritik am Islamismus verboten wird, ist die Demokratie am Ende. Solidarität mit Michel Houellebecq!!!
#Islamismus, #Frankreich, cicero.de/kultur/volksverhet…
Christoph Kletzer retweeted
The Irish feel sympathy for the oppressed people of the world: Russians and Arabs 🤯
Christoph Kletzer retweeted
Je vous présente une des pires pays d'extrême-droite d'Europe, le Danemark 🇩🇰
- 10 points de moins de dépenses publiques que la France
- Excédents budgétaires et dette publique de 29% du PIB
- Licenciement sur simple lettre du patron. Taux de chômage de 3%.
- Pas de SMIC imposé par l'Etat
- Age de départ à la retraite à 67 ans, bientôt 68
- Retraite partition limitée à un minimum vieillesse égal pour tous. Les compléments de retraite se font uniquement par capitalisation
- Environ 48 % de la population a une assurance santé privée
- Politique d’immigration très restrictive : protection temporaire par défaut, regroupement familial difficile, obligation d’intégration réelle
- Résidence permanente seulement après 8 ans + langue + preuve d’emploi soutenu
- Fin du visa automatique pour les migrants au chômage depuis plus de 3 mois
- « Paradigm shift » sur l’asile : protection temporaire par défaut, objectif prioritaire de retour au pays plutôt que d’intégration définitive
- Contrat ordinaire pour l'immense majorité des employés du service public
- Durée de travail limitée à 48h par semaine par l'Etat. Majorité des conventions collectives à 37h par semaine.
- Un des taux d'activité les plus élevés au monde
- Allocation chômage conditionnée à une recherche d’emploi très active : au moins 2 candidatures par semaine, obligation d’accepter des jobs hors de son domaine, risque de perte des droits si on ne prouve pas sa disponibilité
- Interdiction du droit de grève pour certains fonctionnaires
- Impôts sur les bénéfices parmi les plus faibles d'Europe
- Aucun impôt sur la fortune
- Protection de la propriété et justice parmi les meilleures du monde
- Création d’entreprise en 1 jour, entièrement en ligne.
- Classé 7e à l'indice de liberté économique contre 65e pour la France.
- Classé 4e pays de l'OCDE où il est le plus simple de faire des affaires. La France est avant-dernière.
Ça fait peur, très peur. Un véritable enfer néolibéral.
Les heures les plus sombres, à 1 heure de Paris en avion.
This.
Top Chinese models likely trained on NVIDIA chips. Like really likely
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1
78
Christoph Kletzer retweeted
New from @Stripe Economics: employment of young college grads is weaker than normal. Much of the debate is around AI's impact. But evidence is growing that work-from-home is a major factor, which we find in public data & our own Stripe data. 🧵 stripeeconomics.com/p/ai-wor… /1
Christoph Kletzer retweeted
Als Tochter eines türkischen Atheisten, der seine Kinder mit großer Skepsis gegenüber dem Islam erzogen hat, sage ich nur eines: Wer das als sozialen Fortschritt feiert, hat nicht mehr alle Tassen im Schrank.
Christoph Kletzer retweeted
1. More billionaires are associated with less poverty.
2. Increasing billionaires per capita by ten fold lowers poverty by 43 percentage points.
3. Sweden has more billionaires per 10M people than the U.S. (34 vs 30)
4. Correlation is fairly strong (R-squared of 0.6 for two variable cross-section data)
4. Since the graph is only a correlation, what do you think is the cause?
🔗in🧵
Christoph Kletzer retweeted
The problem with all coverage of Polish politics is that there is no objective account to be found in the Western media. From the Western/EU perspective, the Tusk government has the moral high ground because the previous government's judicial reforms were illegal. This is not without substance.
However, all reporting frames events exclusively through this prism. As a result, vital historical context regarding how the transition from communism to capitalism unfolded in practice is lost, as is an understanding of why it remains relevant today. For example, the media often ignore the fact that the opposition believes many communist influences were never properly purged from institutions and thus continue to wield disproportionate and subversive influence.
From the Polish nationalist perspective, only those prepared to defend Polish sovereignty retain the moral and constitutional high ground. This is a particularly sensitive issue for Poles, given the country's history of being "off the map" for 123 years.
Christoph Kletzer retweeted
Great Grok thread on the nature of who does and doesn’t have authority to assess if judges were properly appointed. Turns out everyone is wrong. And nobody has authority. Possibly the case that the whole thing can only be resolved with a new election.
Izabella is right
Believable. This is also why I think journalists shouldn’t refer to economists as objective authorities in their own right just because they claim to be economists. At least 20% of economics is ideological, which is why the school or flavour (aka bias) of economics they champion must be highlighted when citing them.
These leanings are too frequently ignored by the media. Say 80 economists of one distinct school or bias right a letter to complaining about x or y policy. Newspapers cite them at face value and without context, as if they’re an objective body when they’re clearly not.
2
89
Christoph Kletzer retweeted
Believable. This is also why I think journalists shouldn’t refer to economists as objective authorities in their own right just because they claim to be economists. At least 20% of economics is ideological, which is why the school or flavour (aka bias) of economics they champion must be highlighted when citing them.
These leanings are too frequently ignored by the media. Say 80 economists of one distinct school or bias right a letter to complaining about x or y policy. Newspapers cite them at face value and without context, as if they’re an objective body when they’re clearly not.
I've been pointing out since the early 2010s that Thomas Piketty, Emmanuel Saez, and Gabriel Zucman have a bad habit of biasing their data construction to fit their politics.
Whenever they make a suspect call in how to handle stats, it ends up showing that inequality is on the rise. They almost never go in the other direction. The errors are non-random and almost invariably reinforce their political claims about rising income and wealth concentration since 1980.
Piketty did this throughout his book Capital in the 21st Century.
Piketty and Saez did this throughout their famous 2003 paper on inequality in the US.
Piketty, Saez, and Zucman Saez and Zucman did this in their Distributional National Accounts paper that Auten and Splinter dissected.
Saez and Zucman did this in their 2019 book on tax incidence and inequality.
Saez and Zucman did this in their flurry of "studies" and white papers to support Elizabeth Warren's wealth tax proposal, California Prop 40, a new wealth tax in France, and dozens of other similar examples.
A new AI report (linked below) audited 138 papers by Saez and Zucman. It found thousands of errors, 194 of which were serious. A whopping 97% of the serious errors went in the same direction: they showed inequality rising faster or made it higher.
But there's a bigger scandal at play here. Despite these clear biases, Piketty, Saez, and Zucman have enjoyed a stream of praise and prestige within the economics profession. They get expedited and friendly reviews in top econ journals even though a more serious peer review should have caught many of these errors (the Quarterly Journal of Economics is a recurring offender in this regard). Journals regularly give them a pass after the errors in their old work come to light - they're never expected to correct any of it, and just keep on using it as if nothing changed. Their non-peer reviewed data claims are also treated as gospel by major media outlets such as the New York Times and the Washington Post. And they get showered with prizes - Saez and Zucman are both Clark Medal winners from the AEA, and Zucman got his prize in 2023 after evidence of major data shenanigans in his work came to light in 2019.
Until now, those of us who have been trying to correct the record have had to wade through the PSZ trio's cluttered, disorganized, and often non-transparent replication packages (almost all of which are convoluted piles of Excel spreadsheets) to find the problems. I did a line-by-line replication of Piketty and Saez's entire 2003 US income inequality series to use one example, and found it was riddled with errors - almost every single one of which biased their results in favor of their political story.
We had the means to catch this stuff as it was happening, but the economics profession failed to do its job. And it failed precisely because its elite ranks wanted to believe the Piketty-Saez-Zucman story for political reasons.