The AI Rundown by Lightscape Partners – 03/11/24

Good morning, happy Monday, and welcome back to another week of AI news.

Last week in AI:

• New versions of known models was a popular theme last week, as Anthropic and Inflection both announced new LLMs.

• Creatives look away! Madonna announces the visuals for her new tour will be generated using Runway’s generative AI model.

The most important stories:

• A Google engineer is facing charges after stealing sensitive AI documents from his employer, distributing them to China.

• OpenAI claps back at Elon Musk after his lawsuit alleged the company lost its vision, claiming he wanted Tesla to absorb the AI startup.

• TurboTax and H&R Block both learn the word ‘hallucination’ as their new tax advice chatbots gave inaccurate and irrelevant information almost half of the time.

Keep reading to get the Rundown on what happened last week in AI.

Check out the AI conferences happening this month here.

If you’d like a quick refresher on industry terms and the current landscape of AI, skip to the bottom.

Top AI Stories of the Week

Google engineer indicted by the DOJ of stealing AI technology from Google, running his own AI startup in China. Link.

• Linwei Ding is accused of sharing confidential AI technology information with Chinese companies.

• The DOJ has stated Ding ran a Chinese-based startup concurrent to his employment with Google.

• It’s believed Ding a third actor was swiping Ding into the Google campus while he worked over seas on his own startup.

• The engineer stole more than 500 files detailing Google’s AI data center hardware and software.

OpenAI fires back at Elon Musk, releasing old email exchanges. Link.

• In response to Musk’s claims the startup abandoned its original mission, the company released emails from Musk expressing his interest in merging the AI company with Tesla as its “cash cow”.

• Musk claimed a merger with Tesla was the only path that would allow OpenAI to compete against Google.

• Musk continues to bash OpenAI on X, claiming “OpenAI is living a lie.” in a tweet later that week.

TurboTax and H&R Block both try their hand in AI tax advice, resulting in inaccurate tax advice almost 50% of the time. Link.

• Both TurboTax and H&R Block included AI chatbots on their tax-prep software this year.

• Both chatbots failed to answer trivial tax questions like: “Where should my child file taxes if she goes to school in another state?”

• The bots both come with a warning to vet every response given.

• The hasty interjection of AI tools onto a platform illustrates the issue with corporations rushing to implement AI solutions without properly training or developing their own solutions.

Software + Hardware

Anthropic introduces Claude 3, its most powerful chatbot yet. Link.

• The AI startup announced the new model family: Haiku, Sonnet, and Opus. The models are listed in increasing order based on performance and price.

• Anthropic claims the Opus model beats GPT-4 in benchmarks including different education levels of reasoning, code evaluation, and general knowledge.

• Opus and Sonnet are available now, while Haiku will be released shortly. You can chat with Claude 3 Sonnet for free at claude.ai.

Inflection AI Nears GPT-4 performance with new model, Inflection-2.5. Link.

• The new model couples raw capability with Inflection’s signature personality and empathetic fine-tuning.

• Inflection-2.5 is already available to Pi users in browser, on iOS, and Android.

• Inflection’s AI personal assistant, Pi, averages conversation times of 33 minutes with 10% of conversations lasting over an hour.

Image + Video

Madonna becomes one of the first adopters of AI-generated visuals, using Runway to create visual effects on tour. Link.

• Madonna’s celebration tour visuals are all AI-generated using Runway’s model.

• The group was between CGI and generative AI, ultimately deciding the AI-generated visuals were more interesting and appealing.

• The decision comes as more and more creative groups seek job-security from generative AI models.

Pika Labs launched lip syncing to compliment its AI video generation. Link.

• The company known for its generative images tries to differentiate itself with an option to generate sound effects with its videos.

• The sound effects and voices are provided in partnership with Eleven Labs.

Ethics

Microsoft employee claims Microsoft is not doing enough to prevent Copilot Designer producing harmful AI content. Link.

• Shane Jones, a 6-year Microsoft employee, contacted news outlets after testing Copilot Designer in his free time, stating the model was capable of generating disturbing images.

• Jones claims he warned the tech giant about his findings, but has not seen proper action to prevent such misuse.

Cybersecurity researchers develop 'worm' malware to target and spread through AI ecosystems. Link.

• Developed by researchers, the AI-based malware can infect a computer with zero-clicks and read email.

• The “worm” targets AI-powered apps that use tools like ChatGPT and Google’s Gemini.

• The new cyber attack represents a new class, called “zero-click malware”, because the attack is carried out by the automatic actions of the LLM.

Upcoming AI Conferences

IMAGINE AI LIVE ‘24

• Fontainebleau, Las Vegas, NV

• March 27 – 28, 2024

NVIDIA GTC 2024

• San Jose, CA and Virtual

• March 18 – 21, 2024

Get up to speed on the current landscape

• AI Term Glossary

• Current AI Landscape from Sequoia Capital

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].