Michelle Ma

Chatbots are a risky for endeavor for newsrooms (automated and highly-visible), and political chatbots more so. Kudos to Ryan Serpico and all others involved in the development of this fantastic new tool to ask questions about Kamala Harris via San Francisco Chronicle. I like that there's a link to editorial guidelines right there, and a detailed rundown of how it works. And a feedback form! This is a template for anyone looking to do the same at your newsroom. And, if you're looking for AI guidelines before experimenting (as you should), the Poynter Institute has a free template (in the comments)

32

2 Comments

For American tech companies building AI, the past few years have been an unregulated free-for-all. In the race to stuff these expensive new tools into every existing digital product, discussions about potential harms, transparency, and intellectual property have taken a back seat. But yesterday in the EU, the first comprehensive law regulating AI took effect. Here’s what it means. https://lnkd.in/e4QGkZNe

7

I wrote about the new version of ChatGPT's voice assistant, GPT-4o, which can talk, laugh, interpret emotions and even sing. It's clearly modeled on Samantha from "Her," and it means that we are leaving the era of the dry, impersonal AI helper, and entering into much stranger territory. How will society change when AI voices can convince us they're human? Will we fall for our AI assistants, as Theodore does for Samantha? Or will other companies' more detached, less emotional chatbots carry the day? I wrote about the possibilities in The New York Times.

178

18 Comments

Securing AI doesn't mean stifling innovation, I review advice https://lnkd.in/eBQrneGi plus Rashmi Ramesh on AI's data drought, & Tom Field reviews ISMG's North America Midwest Cybersecurity Summit. Moderated by Anna Delaney #AI #cybersecurity #ismg

4

OpenAI’s global influence operation doesn’t run on robots. The ChatGPT maker is expanding its people-powered lobbying, policy, and messaging teams, as it works to shape government rules for artificial intelligence, and to win converts for its tools from Capitol Hill to far-flung foreign countries. The growing OpenAI team soon will include new lobbyists as well as a big name in tech policy, business strategy, and political messaging, Chris Lehane, who joined this week. The artificial intelligence player boosted its footprint on Capitol Hill by 30% early this year, disclosing $340,000 on federal lobbying in the first quarter. More in BGOV: https://lnkd.in/e-Dt5Mia

9

Another big scoop from our AI coverage team. Which I must say, with all due modesty, is on a roll. https://lnkd.in/ggyPJWff

7

Meta's Oversight Board took two cases related to AI-generated explicit images in April. One of the cases was when Meta didn't remove an AI-generated nude image of an Indian public figure despite users reporting it. The Oversight Board has now recommended that Meta change the terminology it uses from “derogatory” to “non-consensual” and move its policies on such images to the “Sexual Exploitation Community Standards” section from the “Bullying and Harassment” section along with other policy changes. However, experts said that Meta needs to look at better user education, more tuned reporting tools, and ditching the 48-hour policy when it comes to addressing issues related to AI-generated explicit images. Inputs from: Aparajita Bharti, Barsha Chakraborty, and Devika Malik Full story: https://lnkd.in/gNhQMBwh #AI #saftey

8

It was great discussing SB 1047, California's controversial AI regulation bill, with legal and policy experts Cecilia Ziniti, CEO of GC AI, Todd O'Boyle at Chamber of Progress, and August Gweon at Covington & Burling. I learned a lot. SB 1047 is causing significant concern for firms like Andreessen Horowitz and Y Combinator, prompting their lobbying efforts. Opponents of the bill highlight three main issues: (1) it places the onus on developers rather than users of AI models, (2) the heavy cost of regulatory compliance discourages smaller developers, and (3) it may stifle open-source AI development and research. If passed, the bill will apply to AI companies, developers, and labs doing business in the state, regardless of where they are based. Governor Newsom's decision will be crucial once the bill reaches his desk. Note: This piece was written before Sen. Wiener responded to a16z and YC. https://lnkd.in/eiYJFDeq Article Link: https://lnkd.in/efiDJyC6

18

4 Comments

🌴 In California, where high-tech innovation and cycles of mass layoffs collide, implementing robust trade secret policies is vital – especially since non-compete rules do not exist as a fallback option. So how do companies attempt to shield their trade secrets at these critical times and what can others learn from them? IAM spoke with specialist lawyers to find out how Californian companies which have to navigate heightened misappropriation risk handle IP protection. The article quotes Irvine-based Jeff Farrow, Michelman & Robinson’s Trade Secrets, Financial Fraud and Executive Disputes chair; San Francisco-based Anjali Srinivasan, a partner at Keker, Van Nest & Peters; Chicago-based Steven Pearlman, partner in Proskauer’s Labor and Employment Law Department; and Los Angeles-based Laura Smolowe of Munger, Tolles & Olson where she co-leads the Trade Secret and Employee Mobility Practice Group – many thanks to you all for sharing your insights! #innovation #layoffs #IPprotection

10

AI systems routinely outperform humans, affirms the Stanford AI Index Report 2024; Should we be worried? There's hardly a day that goes by without someone speculating if AI systems like GPT-4, Gemini 1.5 Pro, Claude 3, and now LLaMA 3 are becoming sentient, the fear being that AI's intelligence, what we better know as artificial general intelligence (AGI) or artificial super intelligence (ASI), will exceed that of the most intelligent humans, making it a sort of Alpha Intelligence that may eventually even enslave humans. While #Nvidia CEO Jensen Huang proposed that AGI would emerge within five years, and Ben Goertzel foresees it happening in just three years, Elon Musk predicts that by the end of 2025 or early 2026, AI will be smarter than any single human, and probably smarter than all humans combined by 2029. Meta CEO Mark Zuckerberg, too, has announced that his company is entering the race to make an AGI but did not provide any timeline. These are all credible voices and we must pay attention. But one also needs to pay heed to equally-accomplished voices like those of Yann LeCun, Andrew Ng, and Gary Marcus, who have begged to differ and have consistently tempered expectations in the context of AGI. Early this month, even Stanford University's seventh edition of the AI Index report weighed in on the same topic, insisting that AI systems like GPT-4, Gemini, and Claude 3 are "impressively multimodal" and "routinely exceed human performance on standard benchmarks". The report, however, qualifies that current AI technology continues to have limitations and cannot reliably deal with facts, perform complex reasoning, or explain its conclusions. What do you think? LiveMint https://lnkd.in/g8VmStwv

28

A day when people can interact directly with computers using their thoughts could be on the horizon. Several companies, including Elon Musk’s Neuralink, have begun preliminary human trials of brain-computer interfaces - devices that decode the electrical signals in their brain and translate them into digital bits. For the latest episode of The Wall Street Journal's Future of Everything podcast, I spoke with neurosurgeon and Precision Neuroscience co-founder Benjamin Rapoport, a company working on brain-computer interfaces. We talked about how the technology works and how these implants could improve the lives of hundreds of thousands of people who could gain the ability to independently engage with the digital world. Take a listen!

11

1 Comment

How AI firms and investors are funding and backing JD Vance & Trump in order to slash AI regulation. A must read - https://lnkd.in/eu4H9gpK

3

One last 🧶yarn🧶 for the year: AI and quantum technology are two of the buzziest sectors out there, and investors are struggling to sort through fact and fiction about what's possible and what is going on. We took a close look at SandboxAQ, a $6bn startup chaired by Eric Schmidt and run by Jack Hidary, which seeks to fuse those futuristic technologies. We found some curiosities, for example, some of SandboxAQ's claimed customers that told us they weren't anymore, or were only involved in testing the company's products. SandboxAQ's first few years as a standalone firm hasn't all been smooth sailing, and its fate rests upon Hidary, who is a bit of a jack of all trades. After dropping out of Columbia University in 1991, Hidary has launched numerous ventures, such as dot-com-bubble era website acquisitions, financial research services startups, lobbying for electric vehicle adoption, a long-shot run at New York mayor, podcasting aggregation, and pursuits in philanthropy, robotics, solar panels, biochemistry and curing cancer. Some of those ventures encountered difficulty, or stopped over after a short period. So, we'll see what happens with SandboxAQ. With the talented Cory Weinberg https://lnkd.in/e99u3FY4

21

1 Comment

As the issues between Alphabet and the US Justice Department continue to play out, it got me thinking about the scale of the tech giant's products/services. The chart below highlights the percent of US adults who use various Alphabet products/services monthly. More than eight in ten used the Google Search Console the past 30 days while two out of three use the Chrome browser or Gmail. The Summer 2024 MRI-Simmons National Consumer Study tells us there are 258M adults 18+ in the US. Of those people... over 244M or 95% used at least one of the Alphabet products/services listed below. #alphabet #google #googlesearch #gmail #youtube #android #bigtech #googlemaps #chrome #chromebrowser #Androidphone

6

1 Comment

I loved working on this story about AI medical "scribes," tools doctors are using to speed up their note-taking. They're buried by paperwork, which demands either multiple extra hours of daily work, divided attention during appointments, or both. The whole sector screwed up with electronic medical records, one doc told me, and is still paying the price. But: The scribes use LLMs, which mostly live in the cloud and sometimes in the United States, raising some biiiiig questions about access to patient data. #Canada #HealthCare #ArtificialIntelligence #MedicalScribes #Ontario

10

1 Comment

Prompting with many examples (so-called "in-context learning", ICL) looks like a powerful alternative to fine-tuning LLMs. As models get better at handling very long prompts, giving them hundreds or thousands of examples directly can outperform fine-tuning, especially for open-ended tasks. It's a trade-off though - fine-tuning has higher upfront costs but is cheaper at runtime. ICL requires more compute for each query but can adapt on the fly. I expect to see more of this approach as LLMs evolve, either as an alternative to fine-tuning or for faster prototyping and as a precursor to generating training data for fine-tuning. Summary: https://lnkd.in/eMcVVK9f Paper: https://lnkd.in/e5_P46kE Follow THE DECODER - EVERYTHING AI for daily #AI news.

2

NEW: The DOJ is arguing Google needs 'structural' changes to prevent using its products like Chrome, Android and the play store to maintain its monopoly in search. This is one of several proposed remedies the government filed Tuesday, in response to a federal judge's August ruling that Google operates a monopoly in search. Also on the government's remedy wishlist: preventing Google from crawling sites for its AI when it crawls sites for search (currently these can't be disentangled) and limits to Pmax. A judge won't decide what Google's punishment until next year (and Google will likely appeal). Read more here 👇 https://lnkd.in/epnZXZdh

51

2 Comments