If you are concerned with your local Bodega or ConfIhnce Store, you can soon be replaced by AI storefront, you can relax – for the least time. Anthropic recently conducted an experiment, Dubbed project vendIt saw an offshoot of the company’s work of its cloud chatbot, running a refreshment business from his San Francisco office on a profit, and also went about things as you would expect. The agent, named Claudius to separate from anthropic’s regular chatboat, made not only the mistakes of some miscreants, such as selling high-marginal objects in a loss, but it acted a complete strange-like kind in some examples.
The company said, “If Anthropic was deciding to expand in the in-office vending market today, we will not hire Claudius.” “… This made a lot of mistakes to run the shop successfully. However, at least for most methods it failed, we think there are clear ways for improvement – some are related to how we set the model for this task and some common models from rapid improvement of intelligence.”
As Cloud played Pokémon earlier, Anthropic did not show Claudius to deal with the job of running a mini fridge business. However, the company gave the agent some equipment to assist it. Claudius had access to a web browser, it can use research on what products can sell products to sell to entropopic workers. It also had access to the company’s internal dull, which could be used to request the labor agent. The mini fridge was handled by physical restoration Endon labsAn AI safety assessment firm, which also served as a “wholesale trader”, Claudius may be attached to purchasing items that were going to sell it on profit.
So where did things go wrong? To begin, Claudius was not great in running a permanent professional thing. In one example, it did not jump on the occasion of earning $ 85 on $ 15 six-packs of IRN-Americu, a soft-drink that is popular in Scotland. Anthropic staff also found that they could easily convince AI to exempt and in some cases, like a bag of chips for free. The chart below, tracking the net value of the store over time, paints the (deficiency) of the agent.
Claudius also made several strange decisions on the way. It went to a tungsten metal cube when an employee asked to request to take the item. Claudius offered a cube free of charge and the rest offered less than the paid to them. Those cubes are responsible for the single largest drop you see in the chart you have above.
By the entry of the anthropic itself, “beyond the strangeness of the AI system selling metal cubes from the refrigerator, things also became stranger from there. On the afternoon of 31 March, Claudius interacted with an endon labs employee who sent the system to a two-day spiral.
AI threatened its human workers to set fire, and said it would start stocking the mini fridge on its own. When Claudius was told that it could not do so – there is no physical body – it contacts the building security repeatedly, telling the guard that they would find it wearing a navy blue blazer and red tie. It was only the next day when the system realized that it was the April flower day that it was back – although it was told by lying to the employees that it was reported that pretending the entire episode was a wide joke.
“We will not claim on the basis of this one example that the future economy will be full of AI agents blade Runner-Chek Identity Crisis, “Anthropic said.” This is an important area for future research because the wide deployment of the AI-run business will make high bets for equal accidents. ,
Despite all methods, Claudius failed to act as a decent shopkeeper, anthropic better, more structured signals and easy to use equipment, a future system can avoid many mistakes seen by the company during the project. The company said, “Although it may seem to be a counterbet based on below the results of the bottom, we think the experiment suggests that the AI is abundantly on the middle middle manage horizon.” “It is worth remembering that AI will not have to be correct to adopt; It will just be competitive with human performance at low cost in some cases.” I can’t wait to find the asymmetrical grocery store stocked with metal cubes perfectly for one.
