Submitted: Important news from the last few weeks1. September 2023
Submitted: Important news from the last few weeks
Apple for Right to Repair
Apple, previously a vocal opponent of the “Right to Repair,” has now given its full support for the California bill.
In a surprise reversal, the company sent Senator Susan Eggman a letter endorsing the deal. It will give Californians the right to repair their own devices.
If passed, SB 244 would require manufacturers to provide consumers and independent repairers with tools, parts and information for electronics and appliance repairs.
Under the draft, tools, parts, and documentation will be available for three years for products priced between $50 and $99.99, and seven years for products over $99.99. .
The senator first introduced the bill in 2018. He received a unanimous 38-0 vote in the California State Senate last May and now faces a final hearing in the Assembly’s Budget Committee.
If approved by the assembly, it will go to Governor Gavin Newsom for signature.
Why does TikTok direct users from the web to their own shop leitp?
As reported by The Information, TikTok plans to block external e-commerce links, including Amazon, directing users to the TikTok store.
Background: TikTok Shop is expected to make $500 million less this year in the US. This is the result of significant investments in hiring, supply networks and retailer support.
Similar to Amazon’s “Sold by Amazon” program, the TikTok marketplace ships various products from Chinese sellers, including clothing, kitchen appliances, and electronics.
The ban prevents creators from promoting items available on their Amazon stores, reducing their commission income.
The strategy aims to boost TikTok Shop’s relatively modest US sales.
In the US, daily spend on the platform has increased from $500k-$1M in June to now $3-$4M, with projections of over $10M by the end of the year.
In Southeast Asia, the TikTok Shop has seen greater success, with daily gross merchandise volume reaching US$50-60 million and expected to reach US$90 million by the end of the year.
Alibaba’s AI language models increasingly popular
Alibaba Cloud has released two open-source AI models, Qwen-VL and Qwen-VL-Chat, trained on the company’s Tongyi Qianwen (Qwen) large language model.
Qwen-VL and Qwen-VL-Chat can interpret visual data such as text in images and respond to location-based requests by, for example, offering directions or interpreting images of signs.
• Qwen-VL can answer open-ended questions about various images and create captions.
• Qwen VL Chat handles complex interactions, including comparing multiple images, answering questions in multiple rounds, writing stories, and creating images from user-entered photos.
According to Alibaba, the models offer superior image recognition and image understanding performance compared to other open-source language models for large-scale image processing.
• The models are open source, allowing researchers and companies to develop their own AI applications without requiring extensive training.
• Alibaba’s previously launched Qwen-7B and Qwen-7B-Chat models boast over 400,000 downloads, a testament to increasing popularity.
AI – Brain implants help the paralyzed
Brain implants combined with AI technology have helped two women – one paralyzed and the other with amyotrophic lateral sclerosis – communicate again.
The breakthroughs, highlighted in two studies presented in the journal Nature, show major advances in brain-computer interfaces.
• In one study, a paralyzed woman had a thin sheet of silicone implanted with electrodes on the surface of her brain.
• The woman, who couldn’t speak or move her arms and legs for 18 years after suffering a stroke, was able to use her brain signals to control a computer.
• An algorithm translated their brain signals into text at a rate of 78 words per minute, five times faster than previous brain-computer interfaces.
• This allowed her to communicate via a talking avatar.
• In another study, a woman with Lou Gehrig’s disease had sensors implanted in the outer layers of her brain.
• After training, a computer system connected to the sensors translated their brain signals into text at a rate of 62 words per minute.
First AI security summit in November with China?
The UK Government has announced that Bletchley Park, known as the home of the WWII codebreakers, will host the world’s first AI Security Summit on November 1st and 2nd. The summit brings together world governments, leading AI companies and experts to discuss the responsible development of AI and the risks involved.
The purpose of the summit is to address the global threats to democracy posed by AI, including their use in warfare and cybersecurity. On the agenda are discussions about AI ethics, misinformation, biased algorithms, autonomous weapons and the availability of semiconductors.
Leading scientists and executives from AI companies such as Google’s DeepMind, Microsoft, OpenAI and Anthropic are invited.
British Prime Minister Rishi Sunak, who has been a strong proponent of AI, announced the summit in June, stressing the need to balance AI’s opportunities with its risks.
The UK government wants to establish London as a hub for global AI regulation and innovation, helped by OpenAI’s decision to open its first non-US office in the city.
At the Bletchley Park building in nearby Buckinghamshire, World War II codebreakers cracked the Enigma code used by the German government in 1941 to control operations on the Eastern Front.
Amazon Prime Video: AI-enabled this season
Amazon Prime Video is introducing AI-powered features to its Thursday Night Football (TNF) programming this NFL season. In 2021, Amazon secured exclusive rights to TNF, which run from 2023 to 2033, as part of the NFL’s broadcast agreements. The new AI features will be available starting September 14th.
• Defensive Alerts use an internal machine learning model to predict attacks by defenders and highlight interesting players with red balls on screen. The model was trained in 35,000 games.
• Prime Targets, formerly known as Open Receiver, uses a green sphere to identify open receivers for QB passes.
• Fourth Down Territory, an analysis tool, shows when a team should go fourth down and what the probability of success is.
• Field Goal Target Zones displays multiple rows showing the probability of success for field goals at various locations.
• Key plays, powered by AI and machine learning, deliver highlights and defining moments in the game.
Former Meta-Ki researchers founded EvolutionaryScale
Former meta-AI researchers, developers of an AI model for biology, founded EvolutionaryScale, a startup that recently raised over $40 million in funding, according to Forbes sources. Led by former Facebook AI scientist Alexander Rives, the team developed the transformer-based model ESMFold, which can predict the molecular structures of proteins from viruses, bacteria and other microbes.
• Last spring, Meta’s ESMFold team was dissolved as part of major layoffs within the company.
• Eight of the founding members of the unit have now joined EvolutionaryScale.
• According to a pitch document viewed by Forbes, the startup was seeking seed funding to significantly expand its AI model.
• Lux Capital reportedly led the approximately $40 million funding round, which reportedly valued EvolutionaryScale at $200 million, with participation from KI-KK investors Nat Friedman and Daniel Gross.
• EvolutionaryScale’s vision is to create a new model every year.
• From the third year, the focus is on the integration of other biological data types beyond protein structure prediction, including DNA sequences, gene expression and epigenetic data.
• The company still predicts that it could be a decade before biological AI models help in the development of new products and therapies.