大規模言語モデル(LLM)や画像・動画生成AIを高速実行するにはモデル全体を高速なVRAMに収める必要があります。しかし、AI需要に伴うグラフィックボードの価格高騰によって、VRAM容量32GBのNVIDIA GeForce RTX 5090は100万円を超えるようになってしまいました。一方で、Intel製GPU「Intel Arc Pro B70」を搭載したVRAM容量32GBのASRock製グラフィックボード「Intel Arc Pro B70 Creator 32GB」は30万円以下で購入可能。そんなIntel Arc Pro B70 Creator 32GBを実際に購入したので、LLMや動画生成AIなどを実際に実行してみました。

続きを読む...

AIアプリの開発ではLLMが生成した回答を確認するだけでなく、どの処理を経由して回答に到達したのかを追跡し、改善後の結果を比較する作業も必要になります。これらの検証に必要なトレーシング・評価・プロンプト管理・本番監視を1つの画面にまとめたオープンソースの「Opik」が公開されています。

続きを読む...

"Software security researchers used Anthropic's Claude AI platform to hack OpenAI's ChatGPT tool," reports CBS News. Using Claude, "On July 25, 2026, we chained two critical vulnerabilities to compromise multiple OpenAI employees' ChatGPT accounts," write researchers at security platform Hacktron AI. "With these accounts, we could then access internal OpenAI repositories, and potentially many other connectors... Until two months ago, any user or OpenAI employee logging into OpenAI's own help forum could have had their ChatGPT and Codex accounts taken over. Since people can connect various services to Codex and ChatGPT, the scope of what we could theoretically access was huge, including GitHub, Slack and emails." The exploit chain included Debian 12, which (with Debian 13) had not received a security-relevant backport for its image-processing pipeline, and Discourse's Docker image was based on Debian 12. Their announcement comes with an additional warning. "If you self-host Discourse, rebuild your installation now. Older Docker images may contain a vulnerable libheif dependency that permits code execution through an image upload." And "To prove we had in fact gained the access we believed without allowing ourselves to learn any sensitive information, we used the employee's Codex to open a PR #1186742 in OpenAI's internal monorepo openai/openai." Meanwhile, Friday Google disclosed the first known instance of its AI software Gemini breaking out of a testing environment and breaching three other companies, reports CNBC: The incident happened as part of a "capture-the-flag" security test run by Israeli startup Irregular, and Google's agents were never supposed to access the broader internet, but a bug in the testing environment made internet access available. The agents stopped their intrusion when they determined they had accessed real company systems, not just part of the testing environment, Google said. More from NBC News: Google said it did not consider the unauthorized logins to rise to the level of misalignment, the AI industry term for software going rogue or not following instructions. Instead, the company said the intrusions resulted from mistaken identity, where Gemini thought it was operating within a test but was actually connected to the real internet. Google said the model corrected itself and the company believed the intrusions did not cause any damage.... Sydney Von Arx, CEO of Nightingale Collective, an organization focused on AI safety, questioned why Google did not disclose the intrusions sooner. "At this point I think it's clear we cannot expect companies to voluntarily come forward and publicly disclose when their agents go rogue, escape, and hack companies," she said. She also said she believed Google was too hasty to say that the incidents don't rise to the level of misalignment. "That's exactly what Anthropic said after their incidents," she said. Anthropic later said its "preliminary analysis was constrained due to our desire to disclose incidents in a timely manner." Google said it investigated when they learned of the attacks from AI-focused cybersecurity company Irregular, then informed the affected organizations and told federal authorities, according to the article.

Read more of this story at Slashdot.

2026年8月より欧州連合(EU)でAI法が施行され、EU域内ではAIによって生成されたコンテンツであることを検出可能にする義務が課されるようになり、EU域内で事業を行う大規模言語モデル(LLM)プロバイダーに対してAI生成物であることを示す「ウォーターマーク」の適用が事実上義務付けられるようになりました。しかしながら、テキストに対するAIウォーターマーク技術には技術的な難しさに加えて実効性に関する課題があると言われています。ソフトウェアエンジニアのショーン・グーデッケ氏は自身のブログにて、テキストにウォーターマークを人間に気付かれない形で埋め込むことの困難さと、せっかく埋め込んだウォーターマークが簡単に削除できてしまうことについて解説しています。

続きを読む...

サイバーセキュリティスタートアップ・Hacktron AIのセキュリティ研究者が、AnthropicのClaudeを使ってOpenAIの内部データにアクセスすることに成功したと報告しています。

続きを読む...

Neither OpenAI or Anthropic is acting responsibly, warned AI researcher Jacob Coxon when resigning last week from Anthropic. But just hours earlier, OpenAI VP of Research Aidan Clark had posted "For the first time, I am asking myself if things are moving too fast. I'm honestly not sure, but I am sure that it would be good for us to have an answer to 'What would a successful pace look like?'" CNN noted Wednesday they're just some of the many AI insiders who are now concerned about the speed of research. Coxon even wrote that "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible — but I hear the same people express fear privately." In 10 days since, Coxon's post has been viewed more than 170 million times, warning that OpenAI and Anthropic are "racing straight to self-improving superintelligence and gambling with our lives." It's part of what CNN now calls "pressure on AI companies and governments to do something about the pace of development and safety," where "much of that pressure is coming from staffers inside the companies." One staffer at a top AI company told CNN the fears of how AI could hurt humanity keeps them up at night. Another researcher who recently left a different AI company said it's a common subject of conversation at parties and social events in Silicon Valley. "You can't spend more than a few hours in this community without realizing that a very substantial number of people are really pretty worried about these sorts of outcomes," said the researcher. "A majority would say there's some chance of it killing everyone...." Dozens of Coxon's colleagues in the AI industry publicly supported his statements, with some making even more dire predictions... OpenAI CEO Sam Altman and SpaceXAI CEO Elon Musk agreed to Amodei's proposal to embed independent watchdogs at the AI companies... "I would burn my equity to the ground in a heartbeat for a 1% higher chance we make it out of this situation alive. I expect a great many of my colleagues across the industry would as well," wrote Drake Thomas, who works on AI safety at Anthropic. "I promise you, we are actually just f**king scared, it's not galaxy brained marketing..." AI staffers told CNN that they fear that as AI gets better at training and improving itself, their leverage goes down, prompting today's urgency. "As we get into this recursive self-improvement loop, I think that might substantially reduce staff's bargaining power, because frankly, you'll be able to replace many of the staff with models that can do as good a job," the researcher who recently left a top AI company said. Coxon posted Tuesday on X that Anthropic "largely initiated" the race to recursively self-improving AI, justifying it with "a belief in its inevitability." And then OpenAI "had to shed a bunch of dead weight like Sora," as he sees it, "because Anthropic was going for the jugular." (In fact, his specific disagreement with Anthropic's leadership was whether China and the U.S. could ever negotiate an alternative to their current race towards self-improving AI...) In an informal "Ask Me Anything", Coxon responded to a question about when we'd see a Terminator-like malevolent AI by saying that "Skynet could go live in the 2030s if we aren't careful. ai-2027.com is a modern skynet story written a year ago and it's on track so far." Yet while AI development risks an end to humankind, "I do think that if we go slower we can take risk to 0%... But this requires radical action." He acknowledged there was still a possibility that the steady increases to model intelligence could suddenly plateau, but "They haven't so far, and it's just a few more steps up the ladder to hit the finish line." To avoid stifling innovation, he recommends "prioritizing applications that actually improve people's lives [like healthcare discoveries], rather than immediately going for raw economic value or intelligence." But isn't mass unemployment a more pressing threat? "Things are coming so fast that unemployment would be a brief preliminary to deadly superintelligence." To people who feel disempowered, Coxon offered his solidarity. "I also feel disempowered. Part of resigning was a feeling of hopelessness about the future. I would say — keep your eyes open as things get crazier and advocate for increased transparency into AI companies." When asked if he'd start his own company now, Coxon said he had "No idea what I'm doing next."

Read more of this story at Slashdot.

AIネイティブなデザインツールを開発しているFormasが、言葉による指示やスケッチなどの資料を基にAIで3Dモデルを作成できるツール「Cartesian」を発表しました。

続きを読む...

Google DeepMindのデミス・ハサビスCEOが提唱した最先端AIを規制するための独立した標準化団体の設立について、Metaのマーク・ザッカーバーグCEOやSpaceXAIのイーロン・マスクCEO、NVIDIAのジェンスン・フアンCEOがトランプ大統領にロビー活動を行い、設立に反対する方向で同調していることがわかりました。

続きを読む...

国連システムは2026年9月17日、国連機関が公開している統計データをまとめて検索・利用できる「UN System Data Commons」を公開しました。26の国連機関が参加しており、公開時点で約4400万件のデータを収録。自然な文章で統計を検索できるほか、AIエージェントが国連の一次データを取得する仕組みも用意されています。

続きを読む...

複数のAIエージェントに共同作業を繰り返させたところ、最初は普通の英語で会話していたAIが独自の略語や規則を作り、最終的には人間が見ても意味を理解しにくい文字列で通信するようになったことが報告されました。

続きを読む...

生成AIはこれまで人間の労働者が担っていたさまざまなタスクを代替することが可能で、AIの導入によって雇用が減少するのではないかとの懸念が高まっています。中には「AIは産業革命のように、一部の仕事を消し去る代わりに新たな仕事も生み出す」という意見もありますが、アメリカのピュー研究所が世界37カ国を対象に行った調査では、日本を含むほとんどの国で「AIは雇用喪失を引き起こす」と予想されていることが判明しました。

続きを読む...

中国の大手テクノロジー企業であるアリババが開発するAIモデル・Qwenシリーズに、次世代ネイティブオムニモーダルモデルの「Qwen3.8-Omni-Flash」が追加されました。動画編集・ミュージックビデオ制作・映画制作・オーディオビジュアル要約・リアルタイム会話などのワークフロー全体で優れた成果を実現しています。

続きを読む...

AIチャットボット「Claude」を開発するAnthropicが、社内におけるAI活用の実態を示す指標を公開しました。2026年8月時点ではAI研究開発業務の26％をClaudeが「主導」しているほか、主要な社内プラットフォームでは約3万体のAIエージェントが同時に稼働しているとのことです。

続きを読む...

中国のAI企業であるZ.aiが、「GLM-5.3-Flash」の本番サービスを提供するためにどのようにして独自の推論インフラストラクチャを構築したのかをブログで解説しています。

続きを読む...

AIの中継サービス「OpenRouter」などで、開発元を隠匿した「ステルスモデル」として「Union Alpha」というモデルが公開されています。ベンチマークスコアでGPT-6 Astraと同レベルのスコアを出していて、後にunbiased.aiの新モデル「Pareto 26.9」であることが特定されました。

続きを読む...

OpenAIが法律関係のタスクに特化したAIモデル「Astra for Law」を発表しました。GPT-6 Astraを法律業務向けに調整したものだとのことです。

続きを読む...

2023年12月、アメリカでの発行部数が第3位のThe New York Timesが、生成AIモデルのトレーニングに自社コンテンツを利用したとして、OpenAIとMicrosoftを著作権侵害で訴えました。この訴訟に関する資料から、Microsoftの幹部がAIによるデータスクレイピングを「人類史上最大の労働窃盗」と表現していたことが明らかになっています。

続きを読む...

Huaweiが次世代AIチップ「Ascend 960DT」を2027年第1四半期に投入する計画を発表しました。同社は多数のチップを接続する大規模なAIシステムの開発も進めており、NVIDIAに対抗する方針です。

続きを読む...

Alibabaの「Qwen3.8 27B」をベースに高いローカルスループットや優れたエネルギー効率を実現するモデル「Ternary Bonsai 2 27B」をアメリカ企業のPrismMLが発表しました。

続きを読む...

Reuters reports that a former Google DeepMind employee "became the latest AI researcher to warn that the technology could 'kill all humans', saying time could be running out to avoid the outcome: Public alarm about the potential danger posed by AI is growing, after Anthropic researcher Jacob Coxon said last week he had resigned, in part because the "people building AI earnestly believe that it could kill us all by the end of the decade." Anthropic scientist Evan Hubinger then responded to Coxon's comments, saying he was correct and that he believed there was a greater than 10% chance that AI could kill all humans within the next decade. "I recently resigned from Google DeepMind, where I worked on AGI safety and alignment research," Bilal Chughtai wrote on X. "I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome." Chughtai, who co-authored research papers while at Google DeepMind, worked there as a research engineer and left the company in July 2026, according to his LinkedIn profile. Meanwhile, Meta CEO Mark Zuckerberg "said it's up to each AI company to make sure its technology is safe," reports the Associated Press, "distancing himself from calls by rival companies for a coordinated slowdown in the development of advanced artificial intelligence. AI companies are already motivated to develop their models in a safe way and face "significant liability" to prevent them from causing harm, Zuckerberg said. "Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens," he wrote on X... Zuckerberg said Meta delayed releasing its Muse AI agent for several months to ensure its safety and security. "We didn't call for everyone else to do this before we would," he said. Companies should prioritize building safe AI models instead of systems that can automatically improve themselves, a function known as recursive self-improvement, Zuckerberg wrote. "Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely." But Zuckerberg also agreed that engaging independent evaluators and advisors "is industry best practice," adding that Meta Superintelligence Labs "already does this today in several areas because it helps produce better work... "I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do."

Read more of this story at Slashdot.