---
title: 'Deepseek''s New AI Has A Secret'
source: 'https://youtube.com/watch?v=pryCaGv8wuo'
video_id: 'pryCaGv8wuo'
date: 2026-10-28
duration_sec: 341
channel: 'Two Minute Papers'
---

# Deepseek's New AI Has A Secret

> Source: [Deepseek's New AI Has A Secret](https://youtube.com/watch?v=pryCaGv8wuo)

## Summary

DeepSeek released a new AI model, 4.1 Flash 1017, just five weeks after the previous version, claiming significant improvements in software building and terminal use. The video analyzes the model's cost-effectiveness, open-weight availability, and the architectural breakthrough of learning from self-feedback, positioning it as a major step in AI capability at a fraction of the cost of frontier models.

### Key Points

- **New DeepSeek model release** [00:02] — DeepSeek 4.1 Flash 1017 is available, released only 5 weeks after the previous version, suggesting a minor update at first glance.
- **Surprising capability jump** [00:31] — The model shows major improvements in building working software and using the terminal, contradicting initial expectations of a small update.
- **Cost per million tokens** [00:43] — API pricing is just over $1 per million output tokens, 15 times cheaper than Claude Opus 5.5, but actual cost per task is higher due to token inefficiency.
- **Cost per task vs per token** [01:32] — DeepSeek uses 2-3 times more tokens than Claude Opus for the same task, so cost per task is still good but less impressive than advertised.
- **Open weights and hardware cost** [01:47] — Weights are free to download and own forever, but running it requires about $1 million in hardware, making it impractical for home use.
- **Race to distill and downsize** [02:16] — New AI systems trigger a race to create smaller, cheaper versions; DeepSeek distills frontier models, and others distill DeepSeek, leading to laptop-runnable versions within a year.
- **Architecture unchanged, secret revealed** [02:46] — Despite no architectural change, the capability jump is attributed to 'learning from self feedback'—the AI critiques and improves its own work without new training data.
- **Self-feedback example** [03:30] — The AI can code a game, play it, detect issues like broken collision detection, and fix them iteratively, learning from its own experience like a human.
- **Paradigm shift in AI learning** [04:01] — Instead of relying on more training data, AI can learn from its own experiences, which is hard because the AI judges and rewards itself.
- **Competition and pace of improvement** [04:48] — Frontier models and open-weight alternatives are improving rapidly, with intense competition leading to cheaper and more accessible intelligence.

## Transcript

Good news everyone, a new DeepSeek AI system is available. Hooray. So, is this a small or a big jump in AI capabilities? Well, doesn't sound big because the previous version was 4.1 Flash and this one is called 4.1 Flash
1017. I know names. Only about 5 weeks have passed since the last version and they didn't even want to increment the version number. So I guess it's not a big jump, right? [laughter]
Boy, was I wrong when I thought that. Holy mother of papers. Look at that. It builds much better working software and [clears throat] is way better at using
the terminal to operate the computer. And it's really good at other things, too. You can run it from the API for just a bit more than a dollar per million output tokens. Wow. But can that really be? This is 15 times cheaper than
Claude Opus 5.5. Finally, intelligence too cheap to meter. Well, that sounds a bit better than it really is. You see, we always measure dollars per a million tokens, but different AI systems use a
different amount of tokens for the same task. And DeepSeek is not among the best in this. In my independent tests, I saw that it can easily take two to three times more tokens to accomplish the same thing as Claude Opus. So, we should
thing as Claude Opus. So, we should always think in cost per task, not per million tokens. Now, the cost per task for DeepSeek is still quite good, just a bit less good as what is advertised. Now, here is the best part. You can
download and own the weights for free and own it forever. You will need H. Let and own it forever. You will need H. Let me check my notes. Ah, yes, exactly $1
to get the hardware to run it. So, I can't run that myself at home. However, that's not the point. The point is that after a new AI system appears, it starts a race to create something close to as good, but at a fraction of its size and
cost. Deepseek does that to the frontier models, while other systems distill Deep Seek down into smaller and cheaper systems. So yes, first one bazillion
dollars will be needed and maybe less than a year later you find yourself being able to run it on a beefy laptop. But wait, the architecture did not change at all in this new version. So how on earth did they get such a huge
jump in capabilities? Dear fellow scholars, this is two minute papers with scholars, this is two minute papers with Dr. Koa. So, how now? I looked at the language they used in their announcement and cross-cheed it against other real
research to see what they may have meant. No guarantees though. But I think meant. No guarantees though. But I think the key term here is learning from self feedback. What's that? That's the opposite of what everybody is talking
about, which is needing to obtain more training data. This is the opposite. This means that the AI is able to critique its own work and learn from it. That is absolutely amazing. Now, hold on to your papers, fellow scholars, and
imagine that it could code up a racing game. Then it starts playing it and notices, wait a minute, the collision detection does not seem to work. Things go through each other. Well, that's not great. Let's fix that. Compile the new
game and keep playing and improving it. Now there is no acceleration. Fix that. Now there is no acceleration. Fix that. Then visual detail is quite poor. Fix that too. And so it goes. So if you remember one thing about this episode,
let it be this. We keep talking about giving more and more training data to AI systems to learn from. But previous research works have shown that we can create situations where the AI can learn from its own experiences learning like a
human. And maybe this is it coming online in a real product we can own and online in a real product we can own and use. That is incredible and very hard to pull off because the AI essentially judges and rewards itself and it's not
always easy to tell when it is doing well or not. The Frontier AI models are well or not. The Frontier AI models are improving at an incredible pace and the open weights alternatives are right on their heels and all this improvement in
just 5 weeks. I am stunned. Amazing. It is lovely to see that there is intense competition in this space. That is amazing news for all of us fellow scholars. More intelligence for all of us, cheaper.
What a time to be alive. I use Lambda to reproduce AI research papers often in minutes. It's also great to train your own models or fine-tune an existing one. Run inference or text to image or video. Easy peasy. Running a Deepseek chatbot
or agent. Super fast, super reliable. Lambda gives you powerful Nvidia GPUs to run your own experiments. I test ideas from the papers I cover and moments from the papers I cover and moments later results. Love it. Seriously, try
later results. Love it. Seriously, try it out now at lambda.ai/papers.
