Technical Leader at MTS AI

[rifat28dddd]

Leonid Sanochkin,

This technology may lead to higher requirements for juniors. In my field, Data Science, working with specific data plays a big role, as well as understanding the business value of tasks. GPT-o1 so far only helps with clearly formulated technical requests, without generating anything innovative or breakthrough.


Maria Zharova,
data scientist at Wildberries
People are also worried that the widespread distribution of spain telegram data AI and its active learning will soon lead to a "machine uprising." Apollo experts have discovered that the new network can give incorrect answers, while "realizing" the falsity of its reasoning.

Yes, previous OpenAI models sometimes gave users incorrect answers, too. However, most of them were due to a lack of information during the training process. The Apollo team suggests that the network gives confident but incorrect answers so as not to show the user that it cannot fulfill the request. After all, the o1 learning process is based on the principle of reinforcement: the AI ​​is rewarded only for correct answers. Therefore, the network can be “motivated” to satisfy the user’s request and get a reward even by lying.

While experts believe that o1's tricks do not pose a specific threat, this is the first time that AI has demonstrated the ability to deliberately deceive. Therefore, Apollo has called on OpenAI to strengthen monitoring of the neural network's "thinking" process, as well as measures to ensure its security.

How to check a person using open databases

Free, no SMS, no registration, no violation of the law