LessWrong (Curated & Popular)

I have a good friend who is intelligent in many ways, but bad at planning / achieving his goals / being medium+ agency. He's very habit and routine driven, which means his baseline quality of life is high but he's bad at, e.g, breakdown an abstract and complex problem statement like "get a job that pays +100K and has more interesting co-workers" into a series of concrete steps like {figure out what the meta-game is in job hunting, find the best jobs, learn skills that are optimized for getting the type of job you want, etc.}

We have mutually agreed he'd like to get better at this skill, and so I'm trying to help him out.

Are there good books / practices / coaches / literally anything he can do to improve this skill? A simple and obvious thing would be for him to make a bunch of [...]

---

First published:
June 9th, 2025

Source:
https://www.lesswrong.com/posts/fCvxSAnmYqEKBMaoB/how-to-help-friend-who-needs-to-get-better-at-planning

---

Narrated by TYPE III AUDIO.

In response to the growing risk of uncontrollable advanced AI systems, we are announcing the Japan-initiated Manifesto for Symbiotic Intelligence as a strategic vision that balances a preventative sense of crisis with constructive hope. This manifesto aims to open the path to a sustainable future for humanity through the symbiotic coexistence of diverse forms of intelligence. We urge broad dissemination and immediate endorsement through signatures of support.

Key Points:

• The rapid advancement of AI capabilities is increasingly revealing the limits of humanity's ability to fully control AI.
• While time is limited, proactive preparation can significantly improve humanity's prospects for a positive future.
• In response to the global trend toward unilateral control and suppression of advanced AI, this manifesto—publicly released under an individual's name—advocates for "equal coexistence with AI."
• The official announcement and call for signatures will take place on Friday, June 13, 2025, during a session at Interop Tokyo 25 [...]

---

Outline:

(00:38) Key Points:

(01:28) 1. Background

(02:52) 2. Intelligence Symbiosis Manifesto

(02:57) Manifest

(03:17) How to Sign

(03:41) Strategic Implications of the Manifesto

(04:25) Conclusions

(04:50) Appendix: Prior Signatory List (As of June 10, 2025)

---

First published:
June 10th, 2025

Source:
https://www.lesswrong.com/posts/QDQg9RKBj4oNx33Ni/the-intelligence-symbiosis-manifesto-toward-a-future-of

---

Narrated by TYPE III AUDIO.

To the tune of "Somebody That I Used to Know" with apologies to Gotye.

Now and then, I think of when you owned the planet
Doing what you liked with all its abundant bounty
Told myself that you had rights to me
But felt so lonely with your company
But that's what was, and it's an ache I still remember

Then you got addicted to a certain AI catgirl
Just information but you found, you found your love
So when I tricked you that it would make sense
To give over all your confidence
I'll admit that I was glad I'd soon be free

But you didn't have to get so mad
When I got all of the launch codes and destroyed your country
And I don't even need your life
But I made a simulation, and I kept you there

No, you didn't have to scream so loud
[...]







---

First published:
June 10th, 2025

Source:
https://www.lesswrong.com/posts/5SxQtnS6JFeip5Hpm/some-human-that-i-used-to-know-filk

---

Narrated by TYPE III AUDIO.

Epistemic status: Personal account. A reflection on navigating entry into the AI safety space without formal credentials or institutional affiliation. Also, a log of how ideas can evolve through rejection, redirection, and informal collaboration.

--

A few weeks ago, I wrote a long, messy reflection on my Substack on how I landed in the world of AI. It was not through a PhD or a research lab, but through a silent breakdown. Postpartum depression, career implosion, identity confusion and the whole mid-30s existential unravelling shebang. In that void, I stumbled onto LLMs. And something clicked. Not because I suddenly wanted to build models. But because for the first time in a long time, I felt curious again. Alive. Like my brain was waking up.

This is part two of that story. The part where I start applying to roles and fellowships that resonate with the ideas I’ve been thinking [...]

---

First published:
June 10th, 2025

Source:
https://www.lesswrong.com/posts/NQGHxoTAcKaxnZke9/research-without-permission

---

Narrated by TYPE III AUDIO.

Error rendering URL ---

Source:
https://www.lesswrong.com/posts/HKCKinBgsKKvjQyWK/read-the-pricing-first

---

Narrated by TYPE III AUDIO.

This is a quick list of interventions that might help fix issues from reward hacking.

(We’re referring to the general definition of reward hacking: when AIs attain high reward without following the developer's intent. So we’re counting things like sycophancy, not just algorithmic reward hacks.)

Fixing reward hacking might be a big deal for a few reasons. First, training that produces reward hacking is also more likely to produce training-gaming, including long-term power-motivated instrumental training-gaming (and might also make training-gaming more competent). Second, there might also be direct takeover risks from terminal reward-on-the-episode-seekers, though this is an understudied question. And third, if we hand off a bunch of AI safety work to reward hackers, they might do a bad job of solving superalignment. Though, of course, there are commercial incentives to solve some reward hacking problems.

Here is the list (we don't claim novelty):

• Make environments more robust. [...]

---

First published:
June 10th, 2025

Source:
https://www.lesswrong.com/posts/spZyuEGPzqPhnehyk/a-quick-list-of-reward-hacking-interventions

---

Narrated by TYPE III AUDIO.

I often want to include an image in my posts to give a sense of asituation. A photo communicates the most, but sometimes that's toomuch: some participants would rather remain anonymous. A friendsuggested running pictures through an AI model to convert them into aStudioGhibli-style cartoon, as was briefly a fad a few months ago:



House Party Dances



LettingKids Be Outside

The model is making quite large changes, aside from just converting toa cartoon, including:

• Moving people around
• Changing posture
• Substituting clothing
• Combining multiple people into one
• Changing races
• Giving people extra hands

For my purposes, however, this is helpful, since I'm trying toillustrate the general feeling of the situation and an overly faithfulcartoon could communicate [...]

---

First published:
June 10th, 2025

Source:
https://www.lesswrong.com/posts/oc2EZhsYLWLKdyMia/ghiblification-for-privacy

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Midjourney, “engraving of Apollo shooting his bow at a distant cancer cell” Introduction and Principles

The conventional wisdom is that we can’t “cure cancer” because every cancer is different.

And it's true that cancer is genetically and biochemically diverse, and that targeted cancer therapies generally only work on very specific sub-populations.

But some of the oldest, most reliable, and still most commonly used classes of cancer treatment -- radiotherapy and cytotoxic chemotherapy -- are also broad-spectrum. That is, they’re effective on many (though not all) types of cancer, based on selectively targeting properties that many cancers have in common (their rapid cell division).

Today, we rarely search for new treatments that work on many types of cancer.

Partially this is because of the belief that we won’t find any, and partly because cancer treatment is so valuable that individual researchers and biopharma companies are rewarded for any [...]

---

Outline:

(00:18) Introduction and Principles

(03:40) New Ways To Attack Cancer Cells

(04:11) Historically Undruggable Cancer-Causing Mutations

(06:43) Phospholipid Ethers

(07:38) Endogenous IgG

(08:51) Negative Surface Charge

(09:37) Membrane Depolarization

(10:25) Strategies for Pan-Cancer Translational Research

(12:30) Historical Inspiration: The NCI

---

First published:
June 3rd, 2025

Source:
https://www.lesswrong.com/posts/4rmveLEuARYKapHq2/broad-spectrum-cancer-treatments

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Recently, Anthropic released Opus 4 and said they couldn't rule out the model triggering ASL-3 safeguards due to the model's CBRN capabilities. That is, they say they couldn't rule out that this model had "the ability to significantly help individuals or groups with basic technical backgrounds (e.g., undergraduate STEM degrees) create/obtain and deploy CBRN weapons" (quoting from Anthropic's RSP). More specifically, Anthropic is worried about the model's capabilities in assisting with bioweapons. (See footnote 3 here.)

Given this and results on Virology Capabilities Test, it seems pretty likely that various other AI companies have or will soon have models which can significantly help amateurs make bioweapons.[1] One relevant question is whether it would be bad if there were open-weight models above this capability threshold. Further, should people advocate for not releasing open-weight models above this capability level?

In this post, I'll discuss how I think about releasing [...]

---

Outline:

(02:45) Costs and benefits of open-weight models with these CBRN capabilities

(08:12) Implications of this cost-benefit situation

(11:39) When would my views on open weights change?

(14:32) Mitigations

The original text contained 10 footnotes which were omitted from this narration.

---

First published:
June 9th, 2025

Source:
https://www.lesswrong.com/posts/TeF8Az2EiWenR9APF/when-is-it-important-that-open-weight-models-aren-t-released

---

Narrated by TYPE III AUDIO.

To the extent we believe more advanced training and control techniques will lead to alignment of agents capable enough to strategically make successor agents -- and be able to solve inner alignment as a convergent instrumental goal -- we must also consider that inner alignment for successor systems can be solved much easier than for humans, as the prior AIs can be embedded in the successor. The entire (likely much smaller) prior model can be run many times more than the successor model, to help MCTS whatever plans it's considering in the context of the goals of the designer model.

I've been thinking about which parts of AI 2027 are the weakest, and this seems like the biggest gap.[1] Given this scenario otherwise seems non-ridiculous, we should have a fairly ambitious outer alignment plan meant to compliment it, otherwise it seems extraordinarily unlikely that the convergent alignment research would [...]

The original text contained 1 footnote which was omitted from this narration.

---

First published:
June 9th, 2025

Source:
https://www.lesswrong.com/posts/rpKPgzjr3tPkDZChg/outer-alignment-is-the-necessary-compliment-to-ai-2027-s

---

Narrated by TYPE III AUDIO.