Emmanuel Théry

Many replies here assume ASI makes human focus irrelevant. "Just tell it to make money." This misses something fundamental: the world is chaotic. Literally — in the mathematical sense. In a nonlinear system, a single human decision can cascade into outcomes no ASI could have predicted or reversed. The butterfly doesn't need to be smarter than the hurricane. So the real class divide won't be focus vs slop. It'll be: — Those who choose their own goals (and use AI to achieve them) — Those who let AI choose their goals for them The first group doesn't need higher IQ than the machine. They need something the machine fundamentally can't provide: a reason to optimize in the first place. "Make me rich" isn't a goal. It's an abdication. The person who knows WHY they want what they want — and can update that why when reality shifts — will outmaneuver the person with a better model but no purpose.

A lot of folks talk about "escaping the permanent underclass". If AGI pans out, the future class divide won't be based on wealth, but on cognitive agency. There will be a "focus class" (those who control their attention and actually do things) and a "slop class" (those whose reward loops are fully RL-managed by AI)

The mean is not necessarily nice. Regression to the mean assumes preferences aggregate into something coherent. They don't. People express goals that are contradictory, incomplete, emotion-dependent. "Rich and famous" — almost everyone wants that. But it's relative and zero-sum. Your exocortex would compute the centroid of 8 billion contradictions. The hard problem isn't executing collective will. It's that most people can't articulate what they actually want — and even when they can, the aggregate is incoherent. What actually matters is understanding each person's real goal deeply enough to trace what would actually serve it. That's not regression to the mean. That's the opposite — it's teleological reasoning, one purpose at a time.

The best case scenario for humanity is that AI destroys all jobs. I mean literally every single one. VC, politicians, engineers, judges. Here's why: regression to the mean. Humanity as a whole can express preferences and values into an exocortex or hivemind. And then the machines can execute that collective will. The general will emerges from vector spaces, and the fact that AI struggles with eccentric thinking is actually a feature, not a bug.

@Barnbarnb5wv @DaveShapi Not necessarily more humans. The requirement is that the generations succession engine doesn't stall.

@Barnbarnb5wv @DaveShapi Need for human-driven technical innovation (finding new ways to achieve goals) will lower dramatically. Need for teleological innovation (setting new goals better aligned with the pursuit of happiness) will increase dramatically.

Low birth rates are a good thing. Agree or disagree?

Iterations yes, but to maximize results towards which goals? If the generation of our great-grand parents was still alive and driving the iterations, they'd be optimizing for goals that'd seem quite odd to us today. I'm not saying it'll necessarily be the case, but I'm quite certain machine iterations can be trapped by bad ideology.

@DaveShapi I actually think it'll slow innovation down. People don't really change. They die. And they end up being replaced by a new generation. That's been one or the most powerful adaptation mechanisms of mankind since the beginning.

Full systematic review: 28 pages, 33 sources, 2025 literature. DM me for the report if you're building/investing in AI. The scaling era isn't dead. But the pure scaling thesis is. 4/4

What's winning: • MoE: 5-10× memory reduction • Quantization: 2.9-4.4× speedup • Test-time compute: Pareto-optimal • Hybrid architectures: Long context, low cost Consensus: ECONOMICALLY SUPERIOR The pivot is happening. 3/4

Hot take that's actually cold in academia: Pure AI scaling has hit economic limits. My agent analyzed 33 papers from 2025. The "just scale it" position is now MINORITY in scientific literature. Here's what the research shows 🧵 1/4

Merci @AlinePicardBFM pour cette opportunité de présenter nos #Kins ! bfmtv.com/lyon/replay-em… via @BFMLyon

@AndrewYNg I agree with your analysis. At @digitalkin_ai we always thought the foundation models' layer would soon be commoditized. Much more value will be created where this abundant intelligence will be applied to real specific business cases, in the agentic layer.

The buzz over DeepSeek this week crystallized, for many people, a few important trends that have been happening in plain sight: (i) China is catching up to the U.S. in generative AI, with implications for the AI supply chain. (ii) Open weight models are commoditizing the foundation-model layer, which creates opportunities for application builders. (iii) Scaling up isn’t the only path to AI progress. Despite the massive focus on and hype around processing power, algorithmic innovations are rapidly pushing down training costs. About a week ago, DeepSeek, a company based in China, released DeepSeek-R1, a remarkable model whose performance on benchmarks is comparable to OpenAI’s o1. Further, it was released as an open weight model with a permissive MIT license. At Davos last week, I got a lot of questions about it from non-technical business leaders. And on Monday, the stock market saw a “DeepSeek selloff”: The share prices of Nvidia and a number of other U.S. tech companies plunged. (As of the time of writing, some have recovered somewhat.) Here’s what I think DeepSeek has caused many people to realize: China is catching up to the U.S. in generative AI. When ChatGPT was launched in November 2022, the U.S. was significantly ahead of China in generative AI. Impressions change slowly, and so even recently I heard friends in both the U.S. and China say they thought China was behind. But in reality, this gap has rapidly eroded over the past two years. With models from China such as Qwen (which my teams have used for months), Kimi, InternVL, and DeepSeek, China had clearly been closing the gap, and in areas such as video generation there were already moments where China seemed to be in the lead. I’m thrilled that DeepSeek-R1 was released as an open weight model, with a technical report that shares many details. In contrast, a number of U.S. companies have pushed for regulation to stifle open source by hyping up hypothetical AI dangers such as human extinction. It is now clear that open source/open weight models are a key part of the AI supply chain: Many companies will use them. If the U.S. continues to stymie open source, China will come to dominate this part of the supply chain and many businesses will end up using models that reflect China’s values much more than America’s. Open weight models are commoditizing the foundation-model layer. As I wrote previously, LLM token prices have been falling rapidly, and open weights have contributed to this trend and given developers more choice. OpenAI’s o1 costs $60 per million output tokens; DeepSeek R1 costs $2.19. This nearly 30x difference brought the trend of falling prices to the attention of many people. The business of training foundation models and selling API access is tough. Many companies in this area are still looking for a path to recouping the massive cost of model training. Sequoia’s article “AI’s $600B Question” lays out the challenge well (but, to be clear, I think the foundation model companies are doing great work, and I hope they succeed). In contrast, building applications on top of foundation models presents many great business opportunities. Now that others have spent billions training such models, you can access these models for mere dollars to build customer service chatbots, email summarizers, AI doctors, legal document assistants, and much more. Scaling up isn’t the only path to AI progress. There’s been a lot of hype around scaling up models as a way to drive progress. To be fair, I was an early proponent of scaling up models. A number of companies raised billions of dollars by generating buzz around the narrative that, with more capital, they could (i) scale up and (ii) predictably drive improvements. Consequently, there has been a huge focus on scaling up, as opposed to a more nuanced view that gives due attention to the many different ways we can make progress. Driven in part by the U.S. AI chip embargo, the DeepSeek team had to innovate on many optimizations to run on less-capable H800 GPUs rather than H100s, leading ultimately to a model trained (omitting research costs) for under $6M of compute. It remains to be seen if this will actually reduce demand for compute. Sometimes making each unit of a good cheaper can result in more dollars in total going to buy that good. I think the demand for intelligence and compute has practically no ceiling over the long term, so I remain bullish that humanity will use more intelligence even as it gets cheaper. I saw many different interpretations of DeepSeek’s progress here in X, as if it was a Rorschach test that allowed many people to project their own meaning onto it. I think DeepSeek-R1 has geopolitical implications that are yet to be worked out. And it’s also great for AI application builders. My team has already been brainstorming ideas that are newly possible only because we have easy access to an open advanced reasoning model. This continues to be a great time to build! [Original text: deeplearning.ai/the-batch/issu… ]

@UBC4ai Come test our Literature reviews Kins at DigitalKin.ai

The arrival of @deepseek_ai R1 as an open source frontier model matching o1 and Sonnet 3.5 capabilities confirms what we saw when we founded @digitalkin_ai in 2023: models are just the engines. They'll soon become abundant and a commodity. There's no real moat protecting @OpenAI other than being first to the market. This advantage will fade away. The real game is puting those fairly abundant and cheap engines at use to solve real complex issues in the real world. That's what we allow you to do if you come to @digitalkin_ai 👍

Soon, anyone will be able to delegate standard digital tasks to decent AI agents at a very low cost. But in high-stakes business situations, only the best wins. That's worth millions. And that's @digitalkin_ai's mission: help enterprises incorporate their unique expertises into their own custom agents to win their most critical battles.

@CJHandmer So rare. So true.

But it's so good in theory!