The next generation of AI models is rumoured be remarkable. But for it to stun the world as ChatGPT did in 2022, fundamental breakthroughs may be needed: https://econ.st/3VZvDxn
Illustration: Daniel Zender
目前，越大越好，所以微軟的預算
2023.09
AI 101: While some students will invariably be more interested in AI than others, understanding the fundamentals of how these systems work is becoming a basic form of literacy—something everyone who finishes high school should know. At the start of the new school year, here are MIT Technology Review’s six essential tips for how to get started on giving your kid an AI education. Read more from Rhiannon Williams and me here.
Chinese AI chatbots want to be your emotional support
What is Chinese company Baidu’s new Ernie Bot like, and how does it compare to its Western alternatives? Our China tech reporter Zeyi Yang experimented with it and found that it did a lot more hand-holding. Read more in his weekly newsletter, China Report. (MIT Technology Review)
Inside Meta’s AI drama: Internal feuds over compute power
Meta is losing top talent left, right, and center over internal feuds about which AI projects are given computing resources. Of the 14 researchers who authored Meta’s LLaMA research paper, more than half have left the company. (The Information)
Google will require election ads to disclose AI content
Google will require advertisers to “prominently disclose” when a campaign ad “inauthentically depicts” people or events. As the US presidential election looms closer, one of the most tangible fears around generative AI is the ease with which people can use the technology to make deepfake images meant to mislead people. The changes will come into effect from mid-November. (The Financial Times)
Microsoft says it will pay for its clients’ AI copyright legal fees
Generative AI has been accused of stealing authors’ and artists’ intellectual property. Microsoft, which offers a suite of generative AI tools, has said it will pay up if any of its clients are sued for copyright violations. (Microsoft)
A buzzy AI startup for generating 3D models used cheap human labor
The Mechanical Turk, but make it 3D. Kaedim, a startup that says it uses machine learning to convert 2D illustrations into 3D models, actually uses human artists for “quality control,” and sometimes to create the models from scratch. (404 media) |
|
。。。。
Google DeepMind said it trained an artificial intelligence that can predict which DNA variations in our genomes are likely to cause disease—predictions that could speed diagnosis of rare disorders and possibly yield clues for drug development.
----
Alphabet-owned Google said on Thursday it would consolidate teams that focus on building artificial intelligence models across its Research and DeepMind divisions in its latest push to develop its AI portfolio.
****
你立足地的AI "服務"，不可不知
日本経済新聞（日経新聞） ·
「メタAI」無償開放 7倍賢くSNSとも連携
https://www.nikkei.com/....../DGXZQOGN18D9O0Y4A....../......新たな基盤技術「Llama（ラマ）3」はスマホでも使いやすいように性能を抑え、高速化を両立。眼鏡型端末や頭に装着するゴーグル型のVR端末メタクエストとの連携も進めます。
。。。。洪士灝。Intel Hapag-Lloyd Point
神經型態計算（Neuromorphic Computing）比目前主流的神經網路（neural network ）更加有趣，有遠勝過神經網路的潛力，但技術發展之路更複雜困難。
目前發展神經型態計算的障礙之一是算力，算力不足也是神經網路的發展在20年前卡住好長時間的原因。依據摩爾定律呈幾何級數成長的晶片計算能力，以及利用GPU大規模平行計算能力來解決一般性問題的通用GPU（GPGPU）在2003年之後日趨成熟，讓神經網路達到實用階段而且日益精進。
神經型態計算技術不只複雜度更高，而且計算的形式關聯到時間，例如Spiking Neural Network需要考量神經的傳導時間、脈衝抵達神經元的時間距離和頻率，需要專屬的計算和網路架構，所以沒辦法像神經網路那樣可以全然借助GPU的算力起飛。
這是一個具潛力的領域，我個人認為台灣產學界可以多關注。不應該所有人都一窩蜂去搞大型語言模型，也應該有人去探索新興領域。
。。。。。
【人工智能】前Google AI專家押注未開發的日本市場
https://goo.su/EYrGF5V
東京初創公司Sakana AI由Google Brain 核心成員和人工智能學者創立。這間初創公司獲得日本政府的支持，也跟NTT和Sony 等企業合作，希望開發日本的AI 市場。
#人工智能 #GoogleBrain #AI #SakanaAI
