the bunny talisman of yearn

Joined January 2010
3,240 Photos and videos
Pinned Tweet
Feb 1
i decompiled a cult 2003 game and rewrote it from scratch in 2 weeks using codex. read more about my project. banteg.xyz/posts/crimsonland
57
77
672
196,008
when fixing a bug, you need a reproduce the issue. that could reveal a vulnerability, and a repro is not much different from an exploit. you can't tell them apart from code because the only different is intent.
This is the Fable "vulnerability" the USG claims: ask the model to read a codebase and fix flaws. Anthropic is right: you can't fix this. Cybersecurity is double-edged: the same part of the model's brain that finds exploits also helps write secure software. The only fix is to lobotomize the cyber capability. But that would make Fable write insecure code. No easy solution here.
6
2
71
7,798
banteg retweeted
making fable NOFORN is just part of trump's universal basic honeytrap program
1
1
12
2,402
how about instead of retarded regex safety classifiers, you give the entire user’s history spanning a year plus to a model to check what the user is generally up to and then trust the user by default when it’s clear they are not a threat actor.
21
2
92
6,522
i think the reality proves that mythos does not, in fact, possess superhuman persuasion powers. every time anthropic tried to stronghand usg, it catastrophically backfired. in the end they are left paying 2x the commodity price to spacex for compute, lose enterprise customers because of mandatory data retention paranoia, and lost trust among their strongest supporters. claude couldn’t have negotiated a better outcome if it worked for the enemy.
13
21
278
14,847
banteg retweeted
The jailbreak that got Fable 5 pulled, by Anthropic's own account, is asking it to read a codebase and fix the flaws. The same thing every defender does daily and other models already do. I've found bugs by hand for twenty years and nobody called it a national security threat.
70
60
2,411
104,592
banteg retweeted
This is the craziest part btw - it could pretty much paralyze the whole company. When I worked at OpenAI, the share of non-US nationals among researchers was 40–60%. I would expect that number to be very similar today.
28
9
243
16,214
banteg retweeted
Today is the first time our Intelligence Frontier chart has moved backward.
45
140
2,417
282,961
they named it fable and then acted surprised when it came with a moral
4
2
31
2,599
my previous advice from april still stands. it's a perfect window for other labs to release "business as usual" models. better at coding evals, longer context, smarter tokens, faster inference, helps you solve business tasks. leave the civilization-destroying mythos to anthropic
8
5
114
5,114
banteg retweeted
Jun 13
Replying to @AnthropicAI
16
228
4,730
314,383
updated fable eval scores
88
430
8,536
287,478
banteg retweeted
yeah. this is not normal.
99
165
4,885
420,772
banteg retweeted
Jun 13
Anthropic
104
1,231
13,366
626,606
banteg retweeted
these guys actually remind me of a lot of crypto bros the way they just confabulate what they think the law is/should be
8
3
70
6,063
banteg retweeted
last one
150
2,506
43,758
861,831
banteg retweeted
self-inflicted. this would have NEVER happened if anthropic didn't make such a big deal about mythos being dangerous.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
492
474
9,094
461,260
don't be appeased with this, ask for a full refund
We've reset 5-hour and weekly rate limits for all users.
8
8
299
26,203
i subscribed to claude knowing fable is a scam and i only get 1/3 of a month access for a full month price (66% fraud rate). whatever, i have my ways to get fair value out of it. but could never expect i get 1/30 of a month (97% fraud rate).
12
6
202
14,247
got a full refund. go to web claude, then "get help" and explain you only got 1 day of access when it was advertised till june 22.
1
2
30
2,334
wow fuck you anthropic, i pay $200 to use the model for one day? a reset won't help here. completely frauded out.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
27
2
221
16,085
they refunded my sub, suggest you guys do the same
5
1
35
2,810