please In the loop
What I Know:
従来の知恵では、これによりソフトウェアエンジニアリングが大幅に加速していると述べています。 But is that? A new survey by METR, released last week, is set up to measure the extent to which AI speeds up the work of experienced software developers. The results were very unexpected.
What the research found – METRは、AIの支援の有無にかかわらず、複雑なソフトウェアプロジェクトに取り組んでいる16人の開発者の速度を測定しました。タスクを終了した後、開発者はAIへのアクセスが平均して20％加速したと推定しました。実際、測定により、AIがそれらを約20％減速させたことが示されました。結果は、AIコミュニティで驚きに激しく会いました。 “I was pretty skeptical that this research was worth doing because I obviously thought there would be a significant speedup,” wrote David Rein, a staff member at Metr, in a post on X.
Why did this happen? – The simple technical answer is: LLM today is good at coding, but often isn't enough to intuitively intuitively that developers are answering perfectly in one shot.つまり、自分でコードを書いたばかりの場合よりも時間がかかる場合があります。しかし、この研究の参加者は、さらにいくつかの人間の仮説も提供しました。 “LLMS is a big dopamine shortcut button that might shoot your problem together,” wrote Quentin Anthony, one of the 16 coders who participated in the experiment. “Do you want to hold down the button that has a 1% chance of fixing everything? It's much more fun than a tough alternative.” (It's also easy to get sucked into a social media scroll while waiting for LLM to generate the answer.
What it means for AI
Who should you know:
Jensen Huang, CEO of Nvidia
Huang found himself in today's news after the US government declared about CNN that he “don't need to worry” about the possibility of the Chinese military using market-leading AI chips produced by his company, Nvidia. 「彼らは単にそれに頼ることができない」と彼は言った。 「もちろん、いつでも制限される可能性があります。」
It's missing – Huang was opposed to a policy that saw the US significantly restrict exports of graphics processing units or GPUs to China, in order to hinder Beijing's military capabilities and AI advances. Nvidia claims that these policies simply encouraged China to build its own rival chip supply chain, hurting US companies and even hurting the US economy.
Selfish debate – Of course, Huang would say, as CEO of a company that lost billions of people as a result of blocking its most sophisticated chips to the Chinese market.彼は、ホワイトハウスでの最近の会議でドナルド・トランプ大統領に彼の視点を説得しようとしてきたとブルームバーグは報じた。
in fact… According to a survey by the Georgetown Security and Emerging Technology Center, the Chinese military uses Nvidia chips and has come to its conclusion by analyzing 66,000 military purchase records.昨年、ニューヨークタイムズは、輸出管理が導入されて以来、大規模な闇市場がNvidiaチップを中国に密輸するために生まれてきました。
AI Action
Instead of reading the paper, many human researchers simply upload them to Claude and chat with their assistants to distill the key findings. 「私は論文の読み方の習慣を変えました」と、人類の上級アライメント研究者であるJan Leikeは、今年初めに時間を語りました。 「今、どこで、通常、私はそれらをクロードに入れて、尋ねます：説明してもらえますか？」
明確にするために、Leikeは付け加えます、時にはクロードが重要なものを間違えます。 「しかし、私が論文を読んだだけなら、私は時々重要なものを間違えます」とLeike氏は言います。 “I think the big effect here is that you can read much more papers than before,” he says, which has a positive effect on his productivity. “A lot of the time you're reading a paper is about understanding whether the paper is completely related to what you're trying to do,” he says. 「そして、その部分はとても速いです 1752603199、実際に重要な論文に集中することができます。 ”
What we're reading
