 |
|
|
|||||
In a new report, OpenAI said it found that AI models lie, a behavior it calls scheming. The study performed with AI safety company Apollo Research tested frontier AI models. It found “problematic behaviors” in the AI models, which most commonly looked like the technology “pretending to have completed a task without actually doing so.” Unlike hallucinations, which are akin to AI taking a guess when it doesn’t know the correct answer, scheming is a deliberate attempt to deceive. Luckily, researchers found some hopeful results during testing. When the AI models were trained with “deliberate alignment,” defined as “teaching them to read and reason about a general anti-scheming spec before acting,” researchers noticed huge reductions in the scheming behavior. The method results in a “~30× reduction in covert actions across diverse tests,” the report said. The technique isn’t completely new. OpenAI has long been working on combating scheming; last year it introduced its strategy to do so in a report on deliberate alignment: “It is the first approach to directly teach a model the text of its safety specifications and train the model to deliberate over these specifications at inference time. This results in safer responses that are appropriately calibrated to a given context.” Despite those efforts, the latest report also found one alarming truth: When the technology knows it’s being tested, it gets better at pretending it’s not lying. Essentially, attempts to rid the technology of scheming can result in more covert (dangerous?), well, scheming. Researchers “expect that the potential for harming scheming will grow. Concluding that more research on the issue is crucial, the report said, “Our findings show that scheming is not merely a theoretical concernwe are seeing signs that this issue is beginning to emerge across all frontier models today.
Category:
E-Commerce
There’s a quiet disaster happening in product organizations right now. Companies are hiring armies of people with “product manager” on their business cards, but they’re treating them like project management with better vocabulary. Frankly I see it a lot with enterprise clients. Teams are drowning in tactical decisions and theyre optimizing for activity over outcomes. The result is not ideal: Products that ship on time but solve all the wrong problems. Roadmaps packed with features nobody asked for. Teams that can execute but have no clue why they’re building what they’re building. THE PROJECT MANAGER TRAP This is what most companies get wrong: They hire product managers to manage timelines, coordinate stakeholders, and shepherd features through development. Basically, they want glorified project managers who can speak startup. But that’s not what product management is. Product management is about making strategic decisions under uncertainty. It’s about understanding users so deeply you can anticipate their needs. It’s about saying no to good ideas so you can say yes to great ones. Project managers ask, “How do we build this faster?” Product managers ask, “Should we build this at all?” WHAT REAL PRODUCT MANAGEMENT LOOKS LIKE In the workshops we run with C-suite leaders, this topic comes up constantly. What separates great product managers from the rest? Three things keep coming up: Curiosity over compliance. The best product managers don’t just execute whatever roadmap is handed to them. They question assumptions, dig into user behavior, and push back when the plan doesn’t make sense. Curiosity is one of the top traits we hire for, by the way. Customer obsession over feature factories. Great product managers spend more time with users than stakeholders. They can tell you not just what features customers want, but why they want them and what happens when they don’t get them. There will never be a replacement for direct conversation with users. Strategic thinking over task management. Project managers focus on the how and when. Product managers obsess over the what and why. They’re constantly asking: “What problem are we actually solving? Is this the right problem?” A SIMPLE HIRING TEST Want to know if you’re hiring a truly product-focused individual? Instead of asking candidates to walk through their process, ask them to defend a decision where they killed a feature everyone else wanted to build. Great candidates will light up. They’ll tell you about the time they said no to the CEO’s pet project because the data didn’t support it. Or when they bravely pivoted the roadmap based on a real conversation with a real user. Weak candidates will struggle with this. They’ll probably talk around stakeholder alignment and delivery timelines because its what they’re focused on. THE STRATEGIC DIFFERENCE Companies that get it right understand that product management is a strategic function, not an operational one. Product managers should be thinking three steps ahead, not three sprints ahead. They should be the people who can walk into a room full of executives and say, “I know we planned to build X, but after talking to users, I think we should build Y instead,” and have the conviction to back it up. That requires a completely different skill set than managing Jira tickets and giving progress or status updates in Slack. A BETTER PATH FORWARD Companies serious about product excellence need to rethink how they hire for the product management craft. Stop optimizing for people who can run efficient meetings and start looking for people who can make hard decisions with incomplete information. Because here’s the thing: In a time where everyone can ship quickly (accelerated by the Bolts and Loveables of the world), competitive advantage comes from shipping the right things. George Brooks is CEO and founder of Crema.
Category:
E-Commerce
Teslas door handle design is notoriously terrible, unintuitive, and dangerous. Customers have been saying it for years. After multiple damning reports, the company might finally be doing something about it. On September 10, Bloomberg published a report on Teslas faulty door handles, which included the discovery of more than 140 consumer complaints related to Teslas doors getting stuck, not opening, or otherwise malfunctioning since 2018. Several instances involved scenarios where passengers trapped inside a burning vehicle appeared to be unable to open the doors from the inside. Five days later, the National Highway Traffic Safety Administration (NHTSA) opened an official investigation into Teslas door-opening mechanisms, citing nine similar complaints from Model Y owners. And on September 17, Tesla design chief Franz von Holzhausen told Bloombergs Hot Pursuit! podcast that the company is actively working on redesigning its door opening mechanisms to make them easier to operate in a panic situation. The issue with Teslas doors comes down to a fundamental design flaw: On the inside of the companys vehicles, the electronic and mechanical door opening systems are entirely separate. That means that, if everything is operational, passengers press a button to unlatch the doors electronically. But if the electronic system is down for any reason, passengers have to rely on the mechanical release system insteadone thats located in an entirely different area of the vehicle. In an emergency situation, its an oversight that can have dangerous consequences. Why Teslas doors are an example of bad design Teslas door construction, which includes handles embedded flush with the car, has long faced critique for its unintuitive design and propensity to become completely frozen in winter weather. But recently, more concerning reports have emerged. On October 24, 2024, four friends died in a Tesla Model Y after the vehicle crashed into a barrier on Lake Shore Boulevard in Toronto and caught fire. Reports indicated that, following the crash, the cars electronic opening mechanism stopped working, preventing occupants from exiting the vehicle. As Fast Company reported at the time, the Model Ys doorsas in all Tesla carsrely primarily on electronic controls. If these fail, the next best option are small release mechanisms sandwiched between the door and the door handle, which passengers unacquainted with the models manual may not be aware of. Teslas manual releases are placed differently depending on the vehicle model. In the Model S, for example, manual release cables are located under the carpet below the front of the rear seats. Bloomberg’s investigation also found that, for some earlier iterations of Teslas top-selling vehicles, there were no manual releases for the rear doors. This included versions of the Model 3 sold between 2014 and 2023, as well as some Model Ys sold between 2020 and 2024. The NHTSAs new investigation highlights this lack of a clear mechanical release system as a primary concern. According to its report, the NHTSA received nine Vehicle Owner Questionnaires (VOQs) from owners of 2021 Tesla Model Ys who reported an inability to open their car doors. The most commonly referenced scenarios involved parents placing a child in the backseat before driving, or exiting the drivers seat to pick up their child from the backonly to find that they were unable to reopen the doors to access the vehicle. Four of these VOQs included that the parents resorted to breaking a window in order to reenter the car. Based on its preliminary review, the agency believes that these situations occur when the car has run out of power and doesnt have enough voltage to electronically open its doors, although no VOQs reported a low voltage battery warning before the incident. Although Tesla vehicles have manual door releases inside of the cabin, in these situations, a child may not be able to access or operate the releases even if the vehicles driver is aware of them, the NHTSAs report reads. It adds, Entrapment in a vehicle is particularly concerning in emergency situations, such as when children are entrapped in a hot vehicle. How Tesla plans to solve the problem Now, it appears that Tesla intends to do something about its faulty doors. Von Holzhausen told Hot Pursuit! that the company is currently working on combining the mechanical and electronic door release functions on its vehicles, presumably to make exiting cars more intuitive if their electronic systems shut down. So, in the moment that youre in a panic situation, the muscle memory to go to what you know is right there, von Holzhausen said. You just pull a little bit further on the lever and you have the mechanical release. Von Holzhausen did not share further details on when these changes might roll out, and he didn’t immediately respond to Fast Companys request for comment on what the new systems will look like in vehicles.
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||