The next generation of DeepMind's language model has been released, and it claims to be close to the reading comprehension of a high schooler.
It says the language model, called Gopher, was able to improve its reading comprehension by consuming massive amounts of texts online.
DeepMind claims that its program has over two hundred billion parameters, which is a measure of size and complexity. It falls somewhere between Microsoft and the Megatron, which has 530 billion parameters, and Openai's GPT-3, which has 175 billion parameters.
A system like this could allow us to summarize information, provide expert advice and follow instructions in a natural language.
The model was able to approach human-level performance on a high school reading comprehension test, according to a paper by DeepMind. The skills of its math and reasoning were not improved.
Generally speaking, the more parameters, the more accurate. Even though the models are large, reading comprehension or harmful stereotypes are proving more difficult to overcome.
Digital assistants and translators have been used with the help of the Gopher. According to DeepMind, a full dialogue with a human was possible in a test.
DeepMind isn't looking to make money.
Koray Kavukcuoglo, DeepMind's vice president of research, told Fortune that that's not the focus right now.
DeepMind published an accompanying paper about the steps researchers took to maintain ethical integrity in an attempt to get ahead of criticism that it used ethnic or gender stereotypes in its algorithm.
The team built a tool called Retrieval-Enhanced Transformer, a massive two trillion word database to cross-reference its sources.
The DeepMind team admitted that research on how language models perpetuate harmful stereotypes is still in its early stages.
Researchers are increasingly turning to stickier problems like the potential for spreading misinformation, as the use of artificial intelligence improves.
Even with all the texts on the internet, helping language models like Gopher read between the lines is proving harder than many researchers admit.
DeepMind has a language that approaches human-level reading comprehension.
Researchers create a neural network to help predict drugs.
Are you interested in supporting clean energy adoption? UnderstandSolar.com can show you how much money you could save by using solar power. Futurism.com may receive a small commission if you sign up through this link.
