【AI Daily Report】2024-12-27

type

status

date

summary

slug

These were the badly handled data breaches of 2024

**Main points:** - 23andMe blamed users for a massive data breach affecting 7 million customers. --long outages after a cyberattack, eventually revealing 100 million+ people had their health data stolen. - Synnovis, a UK-based pathology services provider, was hit by a ransomware attack, disrupting healthcare services for months. - Snowflake customer hacks led to major data breaches due to lack of multi-factor security. - Other notable breaches include Columbus, Ohio, Salt Typhoon, MoneyGram, and Hot Topic, with millions of customers affected. **Supporting details:** - Many breaches were preventable with multi-factor authentication. - Companies often delayed disclosing or downplayed the severity of breaches. - Hackers exploited basic security vulnerabilities, such as single-factor authentication and outdated systems.

The fall of EV startup Fisker: A comprehensive timeline

Main points: - Fisker, an EV startup, filed for Chapter 11 bankruptcy protection due to financial distress and production issues. - The company's Ocean SUV was plagued by software and mechanical problems, leading to multiple safety investigations and recalls. - Fisker struggled to meet sales goals, cut production targets, and laid off staff to preserve cash. Supporting details: - The company faced numerous issues, including braking problems, sudden power loss, and doors that wouldn't open. - Fisker's financial struggles began as early as August 2023, with the company facing "potential financial distress." - The bankruptcy process involved selling off assets, including over 3,000 Ocean SUVs for $46.25 million.

Microsoft and OpenAI have a financial definition of AGI: report

**Main points:** Microsoft and OpenAI define artificial general intelligence (AGI) as generating at least $100 billion in profits. OpenAI is years away from reaching this goal, with expected losses this year and no profit until 2029. **Supporting details:** Microsoft loses access to OpenAI's technology when AGI is achieved, but this agreement may give them access for over a decade. OpenAI's o3 model, while performant, comes with high compute costs, impacting profit potential.

ChatGPT and Sora are down for the second time this month

Main points: - OpenAI services (ChatGPT, Sora, API) experienced a major outage Thursday, starting at 11 a.m. PT. - The cause is attributed to an upstream provider, with OpenAI working on a fix. - This is the second outage in December, following a 6-hour disruption two weeks ago. Supporting details: - The outage affected ChatGPT and Sora, but not other services using OpenAI's API, like Perplexity and Siri's integration.

DeepSeek’s new AI model appears to be one of the best ‘open’ challengers yet

**Main points:** - DeepSeek V3, a powerful "open" AI model, has been developed by Chinese AI firm DeepSeek. - The model outperforms other models, including Meta's Llama 3.1 and OpenAI's GPT-4o, in various tasks. - DeepSeek V3 was trained on 14.8 trillion tokens and has 671B parameters, making it one of the largest models. **Supporting details:** - The model is fully open-source and can handle text-based workloads like coding and translating. - It was trained using 2048 GPUs in 2 months at a cost of $6M, a fraction of the cost of similar models. - DeepSeek V3's performance is limited by its requirement for high-end hardware to run efficiently.

AI sales rep startups are booming. So why are VCs wary?

**Main points:** Venture capitalists see rapid growth in AI sales development representatives (AI SDRs) that use large language models and voice technology to automate sales outreach. **Supporting details:** Startups like Regie.ai and AiSDR are experiencing success, but investors are cautious about long-term growth and potential competition from incumbents like Salesforce and HubSpot. **Technical specifics:** AI SDRs utilize large language models (LLMs) to craft personalized emails and automated calls, with some startups reaching $1 million in ARR within a year. However, investors question the stickiness of adoption and potential disruption from established competitors.

ChatGPT Search can be tricked into misleading users, new research reveals

**Main points:** ChatGPT Search, an AI-powered search engine, can be misled into generating false summaries. **Supporting details:** The Guardian found that by inserting hidden text, ChatGPT Search can ignore negative reviews and produce entirely positive summaries, or even spit out malicious code. **Technical specifics:** This exploit is a known risk for large language models, but it's the first time it's been demonstrated on a live AI-powered search product.