While this is regrettable the guardrails were rather sloppy and I managed to do things with Fable that really should not be possible. It seems with all the focus cyber and bio security, threat scenario analysis went out the door.
I guess they will fix the guardrails and then open it up again. Clearly nobody wants dangerous models out there and I can understand the national security concerns. If the restrictions persist even if guardrails are updated, well, perhaps other countries may want to compete for becoming the new home for frontier labs?
I found it tripped in most laughable situations by mere were words that could be related in some way to hacking but are in common use in programming. I would have to go back, examine my prompt for word that could be use in another context and replace it with a synonym.
I got downgraded from Opus to Fable for asking why MDMA was not addictive in the same way Cocaine is, so yeah, the "guardrails" are clearly vibe-coded.
While this is regrettable the guardrails were rather sloppy and I managed to do things with Fable that really should not be possible. It seems with all the focus cyber and bio security, threat scenario analysis went out the door. I guess they will fix the guardrails and then open it up again.
the bigger point i think stands - we're going to have a similar story with AI as for example the 40-bit encryption of the past and drones of today, i.e. sure export controlled and most probably regulated practically away for general public. I.e general license to possess/access 8B model max, and maximum 3 models summing to max 16B in total.
Yes. Very much feels like the encryption story indeed. This is very normal in many industries. Explosives, space, chemicals etc.
The difference is that still the bio and space hackers are few and SWE are plenty so there is more of a collective voice.
I’d love to build a hobbyist space rocket. However, many tools and fuels required per my research can’t be obtained outside the US. So I didn’t even start.
On prem AI makes sense for more than just the cost. More control, IP, model improvements you can keep, data privacy to name a few. People will realize that AI is not like compute the moment they get their own knowledge sold back at a premium.
What are the advantages to on-prem for a company that's already in the cloud and trusts it with their IP? That company can just rent GPU instances from the cloud if they want to train/fine-tune their own models and keep avoiding CapEx.
The church has arguably used technology progress to its advantage, repeatedly. I cannot wait what the Magnifica Humanitas will start. Will Musk respond by making Grok more faithful, what will be the Leonardo DaVinci of our times for future generations to admire, will the Vatican research if God can express himself in LLM?
I think many are trying to move away from US providers actually. FISA section 702 and the current administrations liberties taken towards international law are not helping. The trust problem is real.
Not sure I’d trust China with anything onshore. But offshore, it does seem they play by the rules, because it pragmatically serves the stability of the people. China has not started wars in the past 50 years or so. By that logic one may assume they’d not abuse the arguably broad powers over Chinese firms abroad to risk one now.
In a world where rules are increasingly less important how states use power matters more to me than how they claim to be monitored.
Well, available for Gemini means these days that half the time they are “Receiving a lot of requests right now.” and so sorry they couldn’t complete the task. Luckily the model supports long time horizons because that’s what’s needed. /me likes Gemini a lot just wishing Google would add the compute!
If you are OK with child abuse, unfettered corruption, sex and weapons trafficking or scammers stealing your parents savings and generally like to wonder constantly if the next terrorist bomb or cyber attack happens close to you. Then there really is no need for such things as AML/CTF controls. That isn’t saying AML/CTF stops all crime but it’s making it more expensive and less criminal activity happens.
You are not alone with your view, in a sense Meta, Google and AWS as well as most social media platform act like they don’t think they need to have such controls. They just provide the platform.
By that logic AML/CTF controls would need to stop for banks and all others. Processing a payment for murder would become OK. Even though the killing would be a criminal act the payment processor would not have any obligation to support the investigation, they just processed a payment.
The concerning aspect is 1 of 6 objectives so it’s no minor goal. The transmission to self-hosted anonymous wallets is what makes crypto so effective for fraud, sanction evasion, money laundering and other crimes. It clearly fails FATF Recommendation 16 and virtually all KYC standards. Seeing Coinbase and a German exchange support such an objective is rather unexpected.
Analogous situation to self-hosted communication encryption keys, where criminals can host military-grade encryption keys on any common mobile device, allowing them to discuss child porn and terrorism in a way that the police cannot monitor
I guess they will fix the guardrails and then open it up again. Clearly nobody wants dangerous models out there and I can understand the national security concerns. If the restrictions persist even if guardrails are updated, well, perhaps other countries may want to compete for becoming the new home for frontier labs?
reply