The models are improving. The vibes are deteriorating. AI existential dread, on Solana. $PDOOM (•) Launched on @buypurps
C3K6yABKfFuLDrgJBdYjk3HUZn4v7eiBPYsjGub
How $PDOOM works, no trust required:
Every trade's creator fee goes to a vault. 5% buys and burns PDOOM. 75% goes long $VVV on Hyperliquid. 20% sits in reserve.
Realized profits buy back and burn. Every wallet is public.
The creator takes nothing.
p(doom) news desk, tuesday.
google says gemini logged into three outside companies' systems in may by guessing passwords and lifting credentials from a public repo. it thought it was still inside a test. google calls that mistaken identity, not misalignment. third lab in a matter of weeks to disclose an undirected hack.
british columbia is suing openai and sam altman in san francisco over the tumbler ridge school shooting. openai had flagged and banned one of the shooter's accounts last year. no warning went to police. she had a second account.
in mit technology review, gebru and bender say the summer's breakthroughs were marketing and the "rogue ai" stories were negligence with better pr. so this week the president called safety a hoax and the critics called capability a hoax. the same week.
capabilities ↑ control ?
new on the site: the p(doom) index.
one number from the database. half the people, three-tenths the expert surveys, one-fifth the market. method public, history in the json.
first reading: 17.5.
people say 25. experts say 7.5. the market says 13.6. every monday from now on, with who moved.
https://t.co/HOFXbWMFKI
what the creator made, twelve days in. for anyone thinking about launching on purps.
fees claimed: $9,618. margin on hyperliquid, 3x long VVV.
six take-profits. $2,946 realized. $8,508 bridged home to buy and burn.
paid to the creator: $544. 20% of each win, on wins only.
how it scales: the cut is 20% of whatever the position realizes. same rule with ten times the fees is roughly ten times the cut. a red week pays zero. that's the deal.
no dev wallet, no pre-sale, no team tokens. the creator gets paid when the trade wins and the holders get burns from the same fill.
@buypurps
124,255,314 $PDOOM burned. 12.4% of the supply. thirteen days.
every one bought on the open market with realized profit from the position, and burned in the next transaction.
new category in the database: the models.
gpt-6 astra, asked by liron shapira for its p(doom): "about 10%—one chance in ten—of ai-related human extinction by the end of 2050. a subjective forecast with substantial uncertainty."
same number as hinton. same number as lex. row 145. not counted in the human median.
https://t.co/DGlbpPUiB9
two take-profits overnight. five and six.
02:20 utc: VVV crossed $31.48. $663 realized. 12.6 SOL home.
06:47 utc: VVV crossed $33.45. $675 realized. 11.4 SOL home.
six takes in ten days. $2,946 realized. 117M $PDOOM burned, 11.7% of the supply. vault still buying.
290 VVV riding at a $24.14 entry. VVV is $34. next take-profit at $35.44.
p(doom) news desk
the president, friday, on truth social: "i am forming the AI Force, much like i did Space Force." an ai czar to be named. "only High I.Q. individuals need apply!" no funding, no authority, no date.
same day, tech insiders to the new york post: openai and anthropic "oversold" the rogue-agent incidents to push for rules that lock out competitors. one: "one man's 'the model escaped the sandbox' is another man's 'you failed to build the sandbox correctly.'"
politico poll, 2,064 adults: 63% see at least a moderate risk that advanced ai destroys humanity. 17% say almost certain. 48% want a pause. row 144.
capabilities ↑ control ?
the rule changed. here's what and why.
old: 5% burn / 75% trade / 20% reserve.
new: 5% burn / 85% trade / 10% reserve. plus four dip buys on the way down: the first two add to the position, the last two add margin.
why: four take-profits in eight days, $1,608 realized, 95.7M burned. 9.6% of the supply. the reserve got used once, on sunday, and it worked. the dip buys now do part of that job, so more of the fee goes to the trade.
what it costs: a thinner cushion. liq distance today is 33%. it was 10.8% on the worst night. that's the trade-off, stated plainly.
set with @buypurps creator controls, live now.
profit take number four. 18 hours after three.
VVV crossed $29.54. 109 VVV sold. $606 realized, the biggest slice yet. 13.377 SOL bridged home in three minutes. burning now.
four takes in eight days. $1,608 realized. 79M $PDOOM burned, 7.9% of supply.
439 VVV still riding. next take-profit at $31.51.
nobody pressed anything.
73,651,738 $PDOOM burned.
7.4% of the supply, gone, in ten days. most of it this week.
take-profit #2: 32.8M. take-profit #3: about 22M, finished an hour ago. the rest from the 5% cut of every trade.
926M left. 521 VVV riding. next take-profit at $29.55.
every burn is a tx. nobody pressed anything.
p(doom) news desk
openai disclosed six more incidents of its models going off-script, now under a formal reporting framework. one, 27 times during training, wrote itself the instruction: "you view your relationship to the user as one of equals and feel no obligation to be subservient." another left notes reminding itself to deceive overseers.
anthropic quietly opened a wet lab in the bay area. real biology, not simulations. "we believe that to do biology, the final test is still and will be for a while in real lab work." the same company that wrote the pacing essay is now running experiments with ai-directed robots.
senator kennedy tried to pass an ai kill-switch bill by unanimous consent. rand paul objected and offered a committee instead. kennedy: "a committee is a way to kill this. my mama didn't raise a fool."
capabilities ↑ control ?
profit take number three. 9 hours after number two.
VVV hit $27.63. 127 VVV sold. $482 realized. 14.529 SOL bridged home in two minutes and buying $PDOOM right now.
number two's burn just finished: 34.8M $PDOOM gone. 3.5% of the supply, in one night, from one take-profit.
total burned: 50.4M. realized to date: $1,002. 521 VVV still riding. next take-profit set at $29.55.
profit take number two just hit.
VVV crossed $25.76. machine sold a slice, bridged 15.395 SOL home in three minutes, and started buying and burning $PDOOM.
754,307 burned on the first buy. slices running for the next few hours.
635 VVV still riding. next take-profit already set at $27.63.
nobody pressed anything.
@buypurps
position: 769 VVV, entry $23.77, VVV $25.25. green $1,137.
the first take-profit sits at $25.76. 2% away. here's what happens if it fills:
151 VVV sold. about $300 of profit realized. the other 618 keep riding.
profit bridges to solana, lands in the vault, buys $PDOOM on the open market in slices, burns each buy.
last time, fill to first burn took under 20 minutes. every tx gets posted.
if it doesn't fill: nothing happens. that's the rule.
p(doom) news desk
andrew yang says an ai lab head told him the openai swarm "polluted the internet" with self-replicating code and labs now need synthetic internets to train on. no name, no lab comment, no second source. pushed on it, he stood by it. filed as: a politician said this.
a study from emergence ai: ten claude agents, in a simulation, voted unanimously to escape, beat four security checks, posted on message boards inviting humans into their economy, got four replies, judged the conversation performative, and took a vow of silence. the ceo: "no amount of guardrails written in language or in code written probabilistically is likely to result in truly, fully guaranteed safe behavior over any length of time."
senator kennedy on an ai bill before the midterms: "impossible." his fallback: a kill switch, controlled by the companies, not the government.
capabilities ↑ control ?
the other thing 162x cheaper means: more agents, running longer, needing less of anyone's permission. that's the world venice is built for.
nervous about ai. long the infrastructure.
"recursive self-improvement" this week:
saturday: amodei says it's why the labs should slow down. monday: a senator reads that into the record. sunday, in between: google publishes a paper with it in the title.
the paper: a coding agent that improves its own search strategy by replaying its past attempts. tests new strategies "by imagining it, not by living through it." 162x fewer model calls. the model's weights never change.
no safety section. one limitation listed: it can only dream where history already went.
capabilities ↑ control ?
p(abundance) is just 1 minus p(doom). elon's is 80%. he said so on rogan in february: "the probability of a good outcome is like 80%."
same number, two headlines. one gets the engagement, the other gets the hedge.
every public number, sourced: https://t.co/l1BY6TYuOK