burkam's Shortform 2024-11-20T21:12:39.256Z

Comment by burkam on Significantly Enhancing Adult Intelligence With Gene Editing May Be Possible · 2024-11-21T07:43:59.571Z · LW · GW

Are you hiring? Do you have a website or list of team members or contact email?

Comment by burkam on burkam's Shortform · 2024-11-20T08:07:41.833Z · LW · GW

Disclaimer: If you have short ASI timelines, this post is not for you. I don't have short ASI timelines and this is not a proposal to build ASI.

Useful background reading: Principles by Ray Dalio. Holden Karnofsky worked at Bridgewater and shaped the EA movement in its image.

 

Isolated Island Plan

1. Most powerful organisations (business or political) in the world rely on secrets to help keep their power.
2. Hacking and espionage are both increasing, making it such that very few organisations in the world can actually keep secrets.
3. Holding a business or political secret between a hundred people is much harder than holding it between five people.
4. Organisations can move much faster if they can iterate and obtain intellectual contributions from atleast a hundred people. Iteration speed is critical to the survival of an organisation. Secrecy slows down organisations.
5. The few organisations that can hold their secrets and do important stuff will become the most powerful organisations in human history.
6. A good way of building such an organisation is to select a few hundred highly trustworthy people, found an island nation with its own airgapped cluster. Complete transparency inside the island, complete secrecy outside. Information can flow into the island (maybe via firewalled internet), but information can't flow out of the island.
7. After many years, if the org has achieved useful shit, they can vote to share it with the outside world.

 

More on hacking

1. TSMC may be able to backdoor every machine on Earth, including airgapped computer clusters of militaries and S&P500 companies.
2. In general, most hardware should be assumed backdoorable. And secure practices should rely on physics, not on trusting hardware or trusting software. The best way to erase a private key from RAM is to pull the power plug. The best way to erase a private key from disk is to smash it into pieces with a hammer. The best way to verify someone else's key is to meet them in person. The best way to disconnect a machine from wireless networks is to put it inside a faraday cage. The best way to export information from a machine is to print plaintext (.txt) on paper and actually read what's on the paper.

 

More on espionage

1. Computer hardware has become cheap enough that basically anyone can perform espionage. You don't need a big team or a lot of money.
2. (This decrease in cost started with the printing press, accelerated with radio and TV, and finally smartphones and global internet. As long as you're not dealing with many days worth of video footage, data collection, storage, transmission and analysis are affordable to millions of people.)
3. Edward Snowden hid NSA documents in an SD card in his mouth. (See: Permanent Record on libgen) A random italian chef could wear a video camera inside his tshirt and obtain HD footage of North Korean weapons brokers. (See: The Mole on youtube)
4. Since anyone can do it, motivations can be diverse. You can do it for the clicks or for money or for the lulz or for the utilons of your preferred ideological side.
5. Anyone can persuade and recruit people from the internet to join their spy org, just as anyone today can persuade and recruit people online for their political group.

 

More on the isolated island plan:

1. Build a secret org on an isolated island to study safe technological progress. Select a few hundred people.
2. Select people from both STEM and humanities background. Ensure initial pool of people is intellectually diverse and was recruited via multiple intellectual attractors. (i.e. not everyone is there because of Yudkowsky, although Yudkowsky fans are accepted as well) The attractors must be diverse enough so the select members don't all share the same unproven assumptions but narrow enough so the selected members can actually get useful work done.
3. DO NOT build a two-tier system where the leaders get to keep secrets from the rest of the island. Complete transparency of leaders is good.
4. Provide enough funding to members to live on the island for a lifetime, and provide complete freedom to pick research agendas including those the leaders and other island members may not like.
5. Economic and military reliance of the island on the US govt is fine. The strategy is to keep secrets and fly under the radar, not wield power and publicly tell everyone you're doing dangerous stuff.
6. Selecting a few hundred people is good (rather than just two or ten) because it allows people lot of freedom to pick their friends, partners and work colleagues. People won't feel as pressured to hide their disagreements with each other and pretend their relationships are fine.

 

Cons of isolated island plan:

1. The island will inevitably end up atleast a bit of an echo chamber, despite attempts to defend against this.
2. Less intellectual contributions from people outside the org means slower iteration speed.
3. Research orgs that are not run top-down have more variable outcomes, as they depend on what large number of people do. Large groups of people are unpredictable and uncontrollable unless you deliberately try to control them.
4. ?

Comment by burkam on burkam's Shortform · 2024-11-20T07:14:39.625Z · LW · GW

Can TSMC backdoor every machine on Earth, including airgapped computer clusters for the US govt and all S&P500 companies?

 

Here's how the backdoor might work:

Maintain a hardcoded set of assembly instructions involved in RSA, AES and ECC encryption.

After every million assembly instructions, read a hundred assembly instructions. Perform set membership to check if these instructions are likely to be RSA, AES or ECC.

If yes, start monitoring every instruction, simulate its running on a separate portion of RAM which is secretly allocated. Find the private key. Encrypt it with Morris Chang's pubkey. Hide this encrypted private key in a secretly allocated portion of RAM and in a secretly allocated portion of disk. (Every TSMC chip knows the addresses where it is hidden)

If the user tries to export their encrypted files from the computer to external disk, also hide the encrypted private key in this data.

Hope and pray the files eventually make it to a non-airgapped TSMC-chip machine. Chip can use some heuristics to check if the machine is "low monitoring". If yes, send the files and private key via internet to Morris Chang's server.