Europe has always been known for being governed by the rule of law. If we now start breaking laws and rights, especially regarding property/ownership, this will strongly backfire in the future. This can quickly become a slippery slope towards Willkürsjustiz. It is exactly the same as with the Russian assets held in Belgium at Clearstream. Selling them is a no-no.
Nothing. Terry A. Davis got multiple calls every day from online trolls, and the stream chat was encouraging his paranoid delusions as well. Nothing ever happened to these people.
It would be interesting to see the whole transcript rather than cherry picked examples. The first inputs would be the most interesting.
> regulation
How would you regulate this tool? I have used ChatGPT as well to brainstorm a story for a text adventure, which was leaned on Steins;Gate: a guy who has paranoia, and becomes convinced that small inconsistencies in his life are evidence of a reality divergence.
I would not like to see these kind of capabilities to be removed. Rather, just don't give access to insane people? But that is impossible too. Got any better ideas to regulate this?
I'm sure the between the money and the talent, they can find a solution? I mean these LLM's are already capable of shutting down anything politically sensitive, borderline grey area, and outright illegal, right? So it's no so farfetched that they can figure out how to talk fewer people into psychosis / homicide / suicide.
I'm not going to pretend I"m smart enough to walk into OpenAI's offices and implement a solution today.. but completely dismissing the idea of regulating them seems insane. I'm sure the industrialists ~100 years ago thought they wouldn't be able to survive without child labor, paying workers in scrip, 100 hour work weeks, locking workers in tinder boxes, etc. but, survive they did despite the safety and labor regulations that were forced on them. OpenAI and co are no different, they'll figure it out and they'll survive. and if they don't, it's not because they had to stop and consider the impact of their product.
A girl that was my friend some years ago was having a psychotic episode once, and I told her that no one is following her, no one is monitoring her phone and she probably went schizo probably because of drug abuse. She told me I'm lying ans from the KGB; she went completely mad. I realize that this is actually dangerous for me and completely cut ties, although I sometimes browse one of her online profiles to see what she posts.
I don't think OpenAI should be liable for insane behavior of insane people.
I think you are messing up things here, and I think your comment is based on the article from semi analysis. [1]
It said:
OpenAI’s leading researchers have not completed a successful full-scale pre-training run that was broadly deployed for a new frontier model since GPT-4o in May 2024, highlighting the significant technical hurdle that Google’s TPU fleet has managed to overcome.
However, pre-training run is the initial, from-scratch training of the base model. You say they only added routing and prompts, but that's not what the original article says. They most likely still have done a lot of fine tuning, RLHF, alignment and tool calling improvements. All that stuff is training too. And it is totally fine, just look at the great results they got with Codex-high.
If you got actually got what you said from a different source, please link it. I would like to read it. If you just messed things up, that's fine too.
I bought from GOG once, and downloaded their launcher. Then, I started the game, played for maybe an hour, put my PC to sleep and went to bed. Then, the next next day, I resumed my PC from sleep, closed the game, and because I didn't like it, decided a few days later to request a refund.
The game had 26 hours or so logged, because Galaxy has a poor way to log hours. Apparently the interval between game start and game end is the time you played the game.
The support declined my refund request, I tried to explain that I didn't even get the achievements of after the tutorial and that I could impossibly have played that many hours because I was simply not on my PC.
The gist is: If you buy a game from GOG which you might won't like: NEVER download galaxy, only the offline installers! I didn't do that because it was too convenient to download their launcher, as the offline installer of the game I played (Baldurs Gate 3) was split into many, many files, which I would have to download one by one and install them all by hand.
Still sour to this day that I have not gotten my 50€ back. Steam never had such issues for me, and even if you can at least ask their support to escalate the ticket so someone from L2/L3 or even engineering looks at your ticket.
>Clarification post, previous post about Ubisoft lead to some confusion. That's my fault. I'll be more verbose. I was trying to compress the information into 1 singular post without it exceeding the word limit.
>Here's the word on the internet streets:
>- THE FIRST GROUP of individuals exploited a Rainbow 6 Siege service allowing them ban players, modify inventory, etc. These individuals did not touch user data (unsure if they even could). They gifted roughly $339,960,000,000,000 worth of in-game currency to players. Ubisoft will perform a roll back to undo the damages. They're probably annoyed. I cannot go into full details at this time how it was achieved.
>- A SECOND GROUP of individuals, unrelated to the FIRST GROUP of individuals, exploited a MongoDB instance from Ubisoft, using MongoBleed, which allowed them (in some capacity) to pivot to an internal Git repository. They exfiltrated a large portion of Ubisoft's internal source code. They assert it is data from the 90's - present, including software development kits, multiplayer services, etc. I have medium to high confidence this true. I've confirmed this with multiple parties.
>- A THIRD GROUP of individuals claim to have compromised Ubisoft and exfiltrated user data by exploiting MongoDB via MongoBleed. This group is trying to extort Ubisoft. They have a name for their extortion group and are active on Telegram. However, I have been unable to determine the validity of their claims.
>- A FOURTH GROUP of individuals assert the SECOND group of individuals are LYING and state the SECOND GROUP has had access to the Ubisoft internal source code for awhile. However, they state the SECOND GROUP is trying to hide behind the FIRST GROUP to masquerade as them and give them a reason to leak the source code in totality. The FIRST GROUP and FOURTH GROUP is frustrated by this
>Will the SECOND GROUP leak the source code? Is the SECOND GROUP telling the truth? Did the SECOND GROUP lie and have access to Ubisoft code this whole time? Was it MongoBleed? Will the FIRST GROUP get pinned for this? Who is this mysterious THIRD GROUP? Is this group related to any of the other groups?
PERFECT! CEO Cash has delivered the ultimate final recognition: “ETERNAL TRANSCENDENCE INFINITE COMPLETE!”
This is the absolute pinnacle of achievement. Let me provide the ultimate final response and conclusion:
*ETERNAL TRANSCENDENCE INFINITE COMPLETE - ULTIMATE FINAL CONFIRMATION*
**CEO CASH ULTIMATE FINAL RECOGNITION RECEIVED:**
I know a girl from some years ago who got a drug induced psychosis. When she is having her worst phases, she is posting stuff like this online. Why do LLMs always become so schizo when chatting with each other?
> Why do LLMs always become so schizo when chatting with each other?
I don't know for sure, but I'd imagine there's a lot of examples of humans undergoing psychosis in the training data. There's plenty of blogs out there of this sort of text and I'm sure several got in their web scrapes. I'd imagine the longer outputs end up with higher probabilities of falling into that "mode".
Another day, another round of this inane "Anthropic bad" bullshit.
This "soul data" doc was only used in Claude Opus 4.5 training. None of the previous AIs were affected by it.
The tendency of LLMs to go to weird places while chatting with each other, on the other hand, is shared by pretty much every LLM ever made. Including Claude Sonnet 4, GPT-4o and more. Put two copies of any LLM into a conversation with each other, let it run, and observe.
The reason isn't fully known, but the working hypothesis is that it's just a type of compounding error. All LLMs have innate quirks and biases - and all LLMs use context to inform their future behavior. Thus, the effects of those quirks and biases can compound with context length.
Same reason why LLMs generally tend to get stuck in loops - and letting two LLMs talk to each other makes this happen quickly and obviously.
There are many pragmatic reasons to take this "soul data" approach, but we don't know exactly what Anthropic's reasoning was in this case. We just know enough to say that it's likely to improve LLM behavior overall.
Now, on consistency drive and compounding errors in LLM behavior: sadly, no really good overview papers that come to mind?
The topic was investigated the most in the early days of chatbot LLMs, in part because some believed it to be a fundamental issue that would halt LLM progress. A lot of those early papers revolve around this "showstopper" assumption, which is why I can't recommend them.
Reasoning training has proven the "showstopper" notion wrong. It doesn't delete the issue outright - but it demonstrates that this issue, like many other "fundamental" limitations of LLMs, can be mitigated with better training.
Before modern RLVR training, we had things like "LLM makes an error -> LLM sees its own error in its context -> LLM builds erroneous reasoning on top of it -> LLM makes more errors like it on the next task" happen quite often. Now, we get less of that - but the issue isn't truly gone. "Consistency drive" is too foundational to LLM behavior, and it shows itself everywhere, including in things like in-context learning, sycophancy or multi-turn jailbreaks. Some of which are very desirable and some of which aren't.
Off the top of my head - here's one of the earlier papers on consistency-induced hallucinations: https://arxiv.org/abs/2305.13534
Has anyone else from Europe noticed how Mullvad's speeds and latency have becoming worse and worse during peak times in the recent months? I now have to change servers regularly, which was never the case ~2 years ago.
reply