Having been simmering for decades, AI has finally taken off and is now dominating top tech firms’ agendas around the world.
In a new development, Microsoft has raised the search limits on its AI search engine, Bing, just a week after imposing them.
Responding to some embarrassing reports of unruly behavior from Bing, Microsoft decided to restrict a user’s daily use of the chat service to five “turns” per session and 50 turns a day as of last Friday.
By Tuesday, the company had received complaints about the restrictions forcing it to increase the usage limits to six turns per session and 60 turns each day. With the new limits, most users will be able to use the new Bing as they normally would, the company announced.
However, it’s unlikely that Meta will maintain its restrictions and regulations as they stand since Microsoft plans to “increase the daily cap to 100 total chats” in the future, according to the company’s blog.
Meta on the other hand, continues to make significant progress with its AI efforts and recently made news over its new AI language generator named LLaMA released on February 24th.
LLaMA differs from ChatGPT or Bing; it is not a system that can be accessed by anyone. In fact, it’s not a system at all, but a quartet of size-varying models used as an instrument for research which Meta intends to share for the sake of “democratizing access in this important, fast-changing field.”
LLaMA models are trained on trillions of tokens and show that it is possible to train sophisticated models using publicly accessible datasets alone, “without resorting to proprietary and incessantly datasets,” according to Meta’s research paper.
In the same vein, Google Quantum AI researchers have experimentally shown that it’s possible to decrease errors in quantum computing by increasing qubit count.
Following this breakthrough, Google and Alphabet CEO, Sundar Pichai, reports that rather than working with physical qubits on their quantum processor one at a time, multiple qubits are being treated as one logical qubit.
These findings mark the first instance in which quantum error correction enhances performance with an increase in qubit numbers, providing insight into achieving logical error rates necessary for computation.
Artificial intelligence, at its most basic level, is human-machine interconnection a la ChatGPT but at an industrial level, it is so much more. Time, deep research and unending experiments can only tell the possibilities.