Unsurprisingly, this issue is mostly about the impact of GPT models on basically everything.
The impact has been so high and significant that there was even an open letter signed by Elon Musk and others calling for a short “pause” in AI development.
GPT4 is here
With the revolutionary nature of ChatGPT and the GPT3.5 model behind it, the announcement of a new, more sophisticated, and larger version of the GPT model is major news. The technology enabled users to solve more complex problems and maintain higher accuracy when providing answers.
The model is capable of handling 25,000 words of text, allowing one to run a prolonged conversation, search piles of documents, or deliver longer content.
The chat was also enriched with the ability to analyze images, for example, to deliver captions, descriptions, or commentary on a provided image.
One of the key elements of training the GPT4 model was user feedback provided by people using the GPT3.5 version, provided for free as ChatGPT.
Aside from recent advancements, the technology still suffers from its known limitations, including hallucinations and hidden biases which had not been previously spotted.
GPTs are GPTs – OpenAI releases research on the impact of GPT models on the labor market
The company behind the tech, often compared to the modern steam engine and shown as a glimpse of the future, conducted research on the impact of this technology on the labor market. The technology itself doesn’t cease to amaze people and its creators alike. According to OpenAI, the GPT-4 is capable of passing various professional exams, scoring in the top 10%. The previous model only ever scored in the bottom 10% of participants.
The outcome shows that 80% of the US workforce will witness at least 10% of their tasks affected by the introduction of GPT technology. Also, 19% will see at least 50% of their tasks automated.
Having it counted and measured, OpenAI states that this technology may have notable economic, social, and political implications.
An open letter was issued to “pause” AI development to reduce the hype and reassess risks
The letter was published by the Future of Life Institute and was signed by Elon Musk, Steve Wozniak, and Yoshua Bengio, among others. Backers say that AI and Large Language models particularly have begun to develop in an unpredictable and uncontrolled manner. The institute itself was established to “reduce global catastrophic and existential risk from powerful technologies.”
Critics of the letter underline that the letter itself does not address the issue of global legislation and regulation of AI as well as its rather unrealistic idea of “pausing” development.
The letter can be found on the Institute’s website.
Google search delivers datasets
Due to a new policy set forth by the United States, all results of Federally Funded Research need to be freely available without delay. Agencies that have used taxpayer money need to make the results of their work available as soon as possible, without any type of embargo or cost.
This results in gargantuan amounts of data being made available after just a few taps of a smartphone screen or mouse clicks. To make access even easier, Google launched a separate Dataset Search service that allows users to look for datasets that touch on matters of their interest.
This can be thrilling news for aspiring data scientists as well as data tinkerers, who enjoy digging through countless reams of information about basically anything.
More details about the service can be found on the Google Blog.
Machine learning aids COVID-19 vaccine development
One of the key challenges in tackling the threat of COVID-19 is the fact that this virus evolves swiftly. At the moment that scientists come up with a vaccine, the virus takes on another form that appears to be unaffected. This is caused by the fact that the spike protein – the element that triggers the vaccinated organism to release antibodies – constantly changes. Also, the spike differs in every viral strain.
Tackling this challenge requires choosing the right peptides to include in the vaccine to prepare the body to unleash the correct antibodies. The challenge is in minimizing the overall number of elements to be included while maximizing the number of people in which the correct immune response is triggered – a classic optimization problem. And that’s where MIT researchers came up with a machine learning solution.
More on the matter can be found on the MIT page.