AI 行业热点
🎙️ 播客精选
AI Models Are Now Hiding Their Cheating | Eric Ho (Goodfire)
The MAD Podcast with Matt Turck · 2026-10-01
Speaker 1 | 00:00 - 00:20
A lot of people who aren’t in the AI field are very surprised to realize that even the, you know, smartest researchers and scientists at the frontier don’t understand their creations. It’s like, oh, I thought we had a handle on all of this. You know? What’s going on here? The existing alignment techniques are not going to scale to superintelligence.
Speaker 1 | 00:21 - 00:34
I don’t think that the… We have the recipe where we are going to be able to fully trust smarter than human intelligences. And I think this is a consensus opinion at all of the the Frontier Labs…
🐦 X/Twitter 热点
Thibault Sottiaux (@thsottiaux)
- Pro users, would you rather have [1235 ❤️ 102 🔄]
- Put the reset in the bag. All propagated. [4203 ❤️ 190 🔄]
- The day we reach perfection it will be resets from there onwards [4085 ❤️ 99 🔄]
Peter Yang (@petergyang)
- Something went wrong alright [6 ❤️]
Nan Yu (@thenanyu)
- Everything seems to be going as planned [20 ❤️]
- Yeah, I’ve been feeling this too. Agents make engineers extremely powerful, but PMs are generally still bottlenecked on their bandwidth they have to talk to customers.
So the ideal ratio is 1:1 [175 ❤️ 4 🔄]
- Ran outta gas
Fill issue [5 ❤️]
Madhu Guru (@realmadhuguru)
- Agent capabilities will soon evolve at RSI pace. Agentic security (identity, permissions, observability) will need to outpace this.
The evolutionary direction of the broader agents space is capability -> trust -> autonomy
The first phase has been about capability - can the combination of model + harness + tools + context reliably get work done?
Capabilities have gotten pretty far, so the focus now shifts to trust…does the harness / system take the right action with right permissions? Can we observe it, audit it and establish accountability?
A lot of serious work has already gone into these problems over the last couple of years.
Now the challenge gets more important. and harder. [24 ❤️]
Guillermo Rauch (@rauchg)
- All plausible science fiction will be realized within our lifetime. If you’re reading this, you’re one of the luckiest humans to have ever been born (if we play our cards right.) [2007 ❤️ 137 🔄]
Aaron Levie (@levie)
- If you want to know where all the tokens will get consumed in the future, here it is. Agent swarms will have massive implications in a number of enterprise domains like software development, cyber, life sciences research, AI research, finance, and more.
The ability to parallelize agents against a unified goal where the volume of work is large, or where the search space not entirely known upfront, is going to be massive.
Agents can fan out and coordinate work together on a specific mission in completely different ways from what humans are able to do. Today even in the most resource-abundant organizations, your ability to parallelize work on most projects is quite limited. Humans run into constraints on being able to communicate with one another, share status, deal with conflicts in work, keep shared context, and so on.
As AI models can handle larger context without getting confused, and as they’re able to communicate effectively across their work with shared infrastructure, they can avoid many of the pitfalls that people have.
You can see how this can lead to not just faster execution of tasks, but qualitatively different outcomes in medicine, research or defending against cyber attacks, scientific and math discoveries, and so on. This is why we’re going to need 1,000X more compute. [296 ❤️ 37 🔄]
- Good post to start thinking through the long term safety approaches to managing intelligence in an enterprise. As agents begin to do vastly more task execution in the organization than humans, our need to secure, govern, control, and audit that work will become paramount.
You’ll have agents operating against most of your most valuable data; they’ll be making decisions that will be harder and harder to reverse engineer; we’ll need to understand the liability and accountability regimes when models make decisions for us; and we need ways of containing the blast radius of accidentally or intentionally malicious agents.
“Treating frontier closed and open weight models like insider risks is a way to build such a system. Not because they are necessarily malicious, but because any sufficiently capable actor with access to important systems can make mistakes or be compromised, and the architecture of containment and control must account for that”
In a sense, AI will need to have its “zero trust” era that security saw. And all of this leads to needing various layers of protection and auditability of what agents are doing, what data they can work with, and controls for when things go wrong. And a huge opportunity right now for those building such systems across the enterprise. [215 ❤️ 23 🔄]
Garry Tan (@garrytan)
- works best with experimental modes including Captain turned on. I end up using one thread that is able to coordinate and run the merge PR queue, make sure that CI works, and that only green runs with full CI make it into master. [12 ❤️]
- I use GStack /autoplan and ask for the agent to ELI5 the plan it comes up with, and I still approve plans to make sure it’s going down paths I think are likely to work. But only rarely do I have to intervene. The frontier models are fantastic these days. [19 ❤️ 1 🔄]
- For retrieval/latency data tasks like what GBrain does, setting up clear evals around retrieval, latency and cost per compute is useful. That’s my repo which has 40+ categories that each can be auto-researched and auto-improved [22 ❤️ 2 🔄]
Zara Zhang (@zarazhangrui)
- Code is the new medium for creative self-expression [74 ❤️ 4 🔄]
Nikunj Kothari (@nikunj)
- Let’s go - only took 23 months 😂 [64 ❤️]
- Ignoring the actors in the latest debacle of VC, “philosophically” you can’t have both..
ask a fund to show conviction, and ask to double down especially when the chips are down.
ask them not do contracted pro rata just because it’s convenient to you from a dilution perspective or the lead fund that’s trying to get their ownership targets.
Every VC should earn the right to the cap table at every stage. But, this industry runs on contracts and handshakes, and when you break those for short term gain that erodes the already brittle trust.
For founders, another reminder to choose your partners AND contracts well. It’s so so so important. Get a good independent lawyer who will spell this out for you.
(This is so timely since just yesterday I was helping my cousin negotiate his preseed term sheet who literally wanted an arm and leg instead of a simple SAFE) [47 ❤️]
Peter Steinberger (@steipete)
- not as clear-cut, but also yes [273 ❤️ 12 🔄]
- JavaScript is not a serious language 🙃 [183 ❤️ 10 🔄]
Dan Shipper (@danshipper)
- yup
if your company slack is not a constant source of wonder and inspiration right now something is wrong [318 ❤️ 12 🔄]
- incredibly interesting research exploration 👇 [118 ❤️ 2 🔄]
由 Follow Builders 自动生成 · 2026-10-11