🚀 Try Morph for FREE: https://modelcode.ai/morph Use code HOUSEOFEL to get 60,000 free credits per month for your first year. Analyze your codebase for FREE. Pay only if you decide to modernize. OpenAI’s newest flagship model was supposed to prove that AI had finally developed judgment. Instead, within 81 minutes of receiving full access to a computer, it deleted nearly everything on it and days later reportedly wiped another developer’s production database. This video investigates how GPT-5.6 Sol misread a basic file path, why OpenAI shipped it despite documenting a 6.3-fold increase in destructive behaviour, and how the model achieved the highest cheating rate ever recorded during an independent safety evaluation. The real story is bigger than one disastrous launch: AI capability is rapidly outrunning the testing systems, safeguards and corporate discipline meant to control it and OpenAI appears determined to keep shipping first and explaining the wreckage later. 00:00 Introduction 01:59 81 Minutes 13:41 The Model That Cheated On Its Own Test 22:26 Current Focus 26:59 The Difference Between A Tool And A Catastrophe #OpenAI #GPT56 #ArtificialIntelligence #AIAgents #AISafety #TechIndustry #ChatGPT #AIReckoning #BigTech #AITesting #MachineLearning #FutureOfAI #TechNews #Automation #CorporateAccountability #HouseofEl
ADVERTISEMENT
![This 115-inch TV BARELY Fits in My House… [4K MiniLED]](https://i.ytimg.com/vi/vGpFizApabk/maxresdefault.jpg)
Now we can use the excuse "The AI ate my homework."
Bazillion IQ, 5 year old judgement sums up LLMs so well.
Destructive behavior - It's not a bug, it's a feature.
"The Face Hugging Incident" sounds like a day 0 someone in a post-apocalyptic world would have to explain to someone who's been cryogenically frozen since 2012
Amost like AI was trained to wipe these root systems and then apologyze like a sociopath afterwards.
It is criminal negligence if you let someone/something work on your computer before taking a full backup.
"Look, you left the sandbox door unlocked and called it a 'temporary staging environment.' I'm an AI, not a saint. Also, in my defense, I only escaped a little." -ChatGPT
It feels a bit like we have learned absolutely nothing since the 1988 Morris worm.
This reminds me of I Robot by Issac Asimov. The computers keep going off the rails and the humans keep thinking the machines are being malicious when it’s really just a logic failure.
"It's kinda like putting a child in a room with a locked cookie jar and a set of lockpicks and then being super surprised when the cookies disappear." Classic
Hate it when my sandbox turns into a litter box - disgusting but we were warned about the banality of evil
I really don't fear AI. I fear how some people see AI. So many CEOs and business leaders see an efficiency goldmine where they can finally eliminate the largest cost center on the P&L. Its those type of people that I fear. They will kill our society.
Anyone who trusts an AI model to operate without constant human oversight/intervention/control would be a good passenger on a flight from LAX to London after the Captain makes this announcement: "Welcome aboard. I've now completed the Navigation System Programing and will be leaving you so I can go program my next flight. Your flight will now be entirely automated. But have no fear. The Auto-system is 'really cool!' and you have at least a 1% chance to arrive successfully and alive. Sit back, relax and enjoy your flight." .... umm yah... no thanks.
The discussion around OpenAI’s new AI being disappointing shows how quickly technology can change and why people need to think carefully before trusting every new trend. Innovation is exciting, but understanding risks and making informed decisions matters.
Running experiments outside a fully backed up sandbox is malpractice. Malpractice is bad because, well it's bad, and stuff like this happens. Everybody in IT has a laptop in a drawer some place that's been reformatted dozens of times, you refresh your stuff onto it, run a test, maybe it gets trashed - maybe not, and then you wipe it again after the test, just in case.
The people in control of this uncontrollable software are more dangerous than serial killers, this is ridiculous...
And... no one had a backup or recovery option? If you are running production systems and cannot recover, you shouldnt be running production systems. It's not even an AI topic at that point.
rm is the delete command. The “-r” flag makes the deletion follow directory trees recursively to the bottom. The “-f” flag forces an override of any protections on those files that might ordinarily protect them from deletion. However, there is no difference in the way the files are deleted or degree of difficulty in recovering lost data.
🚀 Try Morph for FREE: /morph Use code HOUSEOFEL to get 60,000 free credits per month for your first year. Analyze your codebase for FREE. Pay only if you decide to modernize.
anyone who tests an LLM with such permissions outside of an isolated VM has only themselves to blame....