| |
| Here's a look at today's AI briefing: - EU lawmakers pass world's first major AI rules.
- OpenAI to launch Sora to the public this year.
- Google unveils AI agent that performs tasks in video games.
- IBM to cut some marketing staff amid AI push.
Beth
p/beth-duckett | |
| 1 | | EU legislators today granted final approval to the AI Act, paving the way for the world's first major rules governing AI to take effect later this year. The European Parliament approved the rules with 523 votes in favor, 46 against, and 49 not cast. The AI Act should be official law by mid-2024, after EU member states sign off, with its measures rolling out gradually over two years. What it does: - The consumer safety law evaluates AI systems based on risk, with riskier applications facing more scrutiny or outright bans.
- The majority of rules fall on developers of high-risk AI systems.
- The law bans certain AI uses like predictive policing, emotion recognition in schools and workplaces, social scoring systems, and remote AI-powered facial recognition by police in public, except for serious crimes.
- Newer rules require general-purpose AI providers to label deepfakes, disclose training data summaries, and comply with EU copyright law.
- For advanced models like GPT-4, providers must evaluate and mitigate risks, report serious incidents, and reveal their models' energy consumption.
What's next: - The rules will take effect in phases, with a six-month deadline for countries to ban prohibited AI systems.
- Rules governing chatbots take effect a year later, followed by all high-risk systems and others by mid-2026.
- EU countries will form their own AI regulatory bodies for citizen complaints, while Brussels will launch an AI Office to oversee and enforce laws for general AI systems.
- Violations of the Act are capped at $38M or 7% of a company's global revenue, whichever is higher.
The bigger picture: - While the EU is the first to pass comprehensive AI rules, countries like Japan and groups such as the U.N. and G7 are also working on various AI laws and guidelines.
- China finalized its first set of generative AI rules this summer, which require Gen AI providers offering services to the public to obtain licenses. AI-created content is also required to "reflect core socialist values."
- In the U.S., President Biden issued an AI executive order in October, while lawmakers in at least seven states are developing their own AI laws. However, no broad federal rules are in place, while the majority of Americans think AI should be regulated, according to estimates from a 2023 Rethink Priorities poll.
|     | |
| 2 | | OpenAI's Sora, an AI model for generating 60-second videos from text, will launch publicly sometime this year and as early as several months, OpenAI CTO Mira Murati told the Wall Street Journal. Once it launches, Sora users will be able to edit the videos after they are generated, she said. More: - OpenAI first announced Sora in February and continues to add guardrails and fine-tune the model for vulnerabilities.
- Upon launch, Sora will still not have sound in the videos, though OpenAI will "eventually" incorporate audio, she said.
- Sora will also likely be barred from producing images of public figures or overt nudity, Murati said.
- However, OpenAI is working with creators and artists to determine what level of nudity might be appropriate from a creative standpoint.
- "You can imagine that there are creative settings in which artists might want to have more control over that," she said.
Zoom out: - OpenAI trained Sora on licensed videos from OpenAI partner Shuttershock and "publicly available" videos, though Murati didn't specify which platforms.
- She also noted that it's far costlier to generate Sora videos than Dall-E images. But Sora will be optimized to use less computing power when it's released.
| | | |
| |
|
| 3 | | Google DeepMind unveiled SIMA, a "generalist" AI agent that can perform hundreds of tasks across unseen 3D game environments using natural-language instructions. SIMA, which is only in the research phase, can learn how to navigate 3D virtual settings, including those it hasn't seen before. More: - SIMA, or Scalable Instructable Multiworld Agent, comprises multiple models, including those that map language with visuals and a video model that predicts what happens after the agent acts.
- To train and test the AI agent, Google collaborated with eight game developers, including Hello Games, Embracer, Tuxedo Labs, and Coffee Stain.
- They trained SIMA in open-play games, including No Man's Sky, Valheim, and Goat Simulator 3. In No Man's Sky, for example, it performs at roughly 60% of human capacity.
- SIMA, with 600 basic skills like turning left and climbing ladders, could eventually perform more complex in-game functions.
- The eventual goal is to develop the agent to play any game, including more difficult open-world ones, and perform tasks in unfamiliar real-world settings.
| | | |
| 4 | | IBM informed staff on Tuesday that it will cut jobs in its marketing and communications division as the company shifts priority toward AI. The company said the job cuts align with its "rebalancing" effort to match its workforce skills with client needs in AI, hybrid cloud, and others areas. More: - IBM's Jonathan Adashek disclosed the layoffs during a seven-minute meeting, though the exact number of cuts is unclear.
- In its January earnings call, the company said it planned to cut 1.5% of its global workforce.
- CFO James Kavanaugh told Bloomberg that the layoffs would affect about 3,900 jobs but added the company plans to hire in "higher-growth areas."
- IBM said it plans to end 2024 with about the same headcount as a year earlier.
Zoom out: - In May, IBM CEO Arvind Krishna said the company would stop recruiting for some positions that could potentially be replaced by AI, specifically focusing on back-office functions like certain HR roles.
- Krishna predicted that about 30% of these roles, employing around 26,000 employees at IBM, could be replaced by AI and automation within five years.
| | | |
| |
|
| 5 | | 🍋 FRESH FUNDING 🍋 - Bear Robotics, a startup making autonomous AI-powered server robots for restaurants and other venues, raised $60M in Series C funding exclusively from LG Electronics. The investment makes LG the single largest shareholder in the firm.
- Peter Thiel's Founders Fund, former Twitter executive Elad Gil, and other investors provided $21M in funding to Cognitive AI, a startup that can turn a user's prompt into a website or video game.
- Index Ventures led a $20M seed funding round into Adaptive, a new startup making it easier for businesses to train large language models. Investors participating in the round include ICONIQ Capital, Motier Ventures, Databricks Ventures, and HuggingFund by Factorial. The startup was valued at $100M.
For the latest on AI startup funding and more in the VC community, subscribe to our Inside Venture Capital newsletter. | | | |
| 6 | | QUICK HITS *This is a sponsored post. | | | |
 | AI and technology writer | | Beth is a contributing editor and writer of Inside's AI and Tech newsletters. She has written for publications including USA Today, the Arizona Business Gazette, and The Arizona Republic, where she received recognition with a Pulitzer Prize nomination and a First Amendment Award for collaborative reporting on state pension cost increases. You can reach her at Beth.Duckett@yahoo.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
 | UXPin Merge – a UI builder for busy Developers & Designers. Create coded layouts in minutes. | |
|
 | Ortto is a complete platform for powerful marketing automation. Start your 14-day free trial today. | |
|
| |
Comments
Post a Comment