Gemini : The ChatGPT Killer ?

Good morning everyone !

Another interesting week.

We’re getting jaded.

We’re not really surprised anymore by an AI advancement that 2 years ago would seem science-fiction like Deepmind’s Gemini announcement.
 
Same goes for cyber where ransomware is ravaging the business world and millions upon millions of accounts are breached every week. We’re not batting an eye.

It’s getting so bad that I’m of the humble opinion that there’s only one thing that can save us : benign AI.

And to be honest : I also start to think that about humanity in general. AGI , the Singularity, might be at the same time the biggest threat to humanity or the only solution to the gigantic mess we find ourselves in.

Anyway, this is bar talk and it’s only friday morning.

[AI]
Google’s long-anticipated ‘ChatGPT killer‘ is finally here

Google DeepMind just revealed THREE versions of their most capable multimodal AI yet — Google Gemini (aka GPT-4’s biggest competitor).

Gemini comes in three sizes — Ultra for complex tasks, Pro for scaling across a range of tasks, and Nano for efficient on-device tasks.

Gemini Ultra’s performance beats current state-of-the-art results in 30 of 32 benchmarks used in LLM research & development.

With a score of 90%, Gemini Ultra is the FIRST AI model to outperform human experts on the MMLU benchmark.

Gemini has next-generation capabilities such as sophisticated reasoning, multimodality, and advanced coding.

Google also announced Cloud TPU v5p, their most powerful TPU system to date, which Gemini is running off of.

Gemini Pro will be in Google products (Gmail, YouTube, Docs, and more) through Bard starting today.

Bard is now available in English in more than 170 countries here.

Gemini Ultra will be rolling out early next year.

Gemini will also be available directly on-device in Pixel 8, using Gemini Nano.

Like the article linked above suggests this is a big step for Google but not necessarily a giant leap for the field as a whole.

[CYBER]
23andEveryone

In what is another gigantic breach, 23andMe has confirmed a data breach affecting 6.9 million users. Roughly 5.5 million of these users had their genetic matchmaking profiles exposed - yes, we’re talking about the DNA Relatives feature. Another 1.4 million had their family tree profiles accessed. The breach, revealed in an SEC filing and a late-night blog update, was the result of a credential stuffing attack. Essentially, someone got a hold of old login info and went on a digital shopping spree through 23andMe’s data aisles.

Here’s where it gets even more sci-fi: the attacker used the DNA Relatives feature to expand their reach, affecting millions more than the initial 14,000 directly accessed accounts.

Despite the breach, 23andMe’s spokesperson insists their systems are secure, which sounds a bit like locking the barn door after the horse has bolted. Especially since the stolen data was already spotted on the dark web back in October. We’re talking full genetic profiles here - not just names and birthdates, but DNA matches, family names, even profile pictures.

While 23andMe is beefing up security with mandatory two-step verification, this is a giant hit to the company.

[Cyber]
The Week in Cyber

Sellafield Nuclear Site Hacked or Not ? Despite a report by the Guardian alleging such a breach, the UK government has refuted claims that Chinese and Russian hackers infiltrated the Sellafield nuclear site.

Pan-American Life Insurance Group MOVEit Attacks: The Pan-American Life Insurance Group (PALIG), which has joined a list of more than 2,600 affected businesses worldwide, is one of the most recent victims of the MOVEit cyberattacks. These attacks have compromised the data of approximately 77 million users.

Data Breach at RailYatri: The Indian travel business RailYatri experienced an internet data breach in December 2022 that affected over 30 million members. The Have I Been Pwned (HIBP) program has identified this incident.

Ransomware Attack on Tipalti: According to the AlphV group's report on their dark web website, payment software provider Tipalti has confirmed a ransomware attack. Notably, AlphV has threatened to use the stolen data to extort Tipalti's major clients, including Roblox and Twitch.

Windows 10 Extended Security Updates: Microsoft has announced that it will provide three years of paid security updates for Windows 10 users beyond its end-of-support date on October 14, 2025, through the Extended Security Updates (ESU) program.

Harvard University's Controversial Handling of Facebook Studies: Harvard University apparently suppressed a team of researchers investigating Facebook's disinformation tactics after a sizable gift from Meta CEO Mark Zuckerberg. The team was working on the 'Facebook Files,' which were based on internal Meta papers disclosed by whistleblower Frances Haugen.

International Crackdown on Money Mules: In a major operation by Europol, Interpol, and law enforcement agencies from 26 countries, 1,103 money mules have been arrested in a significant move against money laundering operations. Nearly 11,000 money mules and their recruiters have been identified, with the assistance of over 2,800 banks and financial institutions.

Unitronics Hacking Incident by Cyber Av3ngers: The US government has confirmed that the Iranian hacking group Cyber Av3ngers infiltrated water facility equipment across multiple US states. The group exploited vulnerabilities in Unitronics PLCs, often using the default password. US agencies have urged organizations to enhance security measures, including changing default passwords and enabling MFA.

EU Cyber Resilience Act Agreement: The European Council and Parliament have reached an agreement on the Cyber Resilience Act to bolster the security of smart devices sold in the EU. The act mandates vendors to establish vulnerability reporting processes, support products for a minimum of five years, and provide free, automatic security updates. It also requires data encryption for confidentiality and mandates vendors to report any attacks. The act will be effective three years after its formal approval.

[AI]
The Week in AI

ChatGPT Turned 1: Last week , on November 30th - our friend ChatGPT turned exactly one year old. It’s amazing to think about how we work has changed since it’s arrival and how much it will change going forward.

StabilityAI fighting with its investors : Stability AI, a prominent company in the artificial intelligence sector, is currently facing internal challenges. Investors are pressuring the CEO, Imad MTO, to resign due to leadership concerns and the company's precarious financial situation. Although there are rumors about the company seeking buyers, with Cohere and Jasper mentioned as potential acquirers, the CEO has publicly denied these claims. This situation has caught the attention of the broader AI industry, leading to speculation and anticipation of further developments.

DALLE3 now available in MS Paint Microsoft has integrated DALL-E 3, a powerful AI image generation tool, into Microsoft Paint. Users can now create images in Paint using the new "co-create" feature, where they describe their desired image – for example, a humanoid robot holding a basketball – and choose a style, like digital art. This feature uses a credit system for generating images, the details of which, such as credit allocation and refresh rates, are still unclear. Once an image is generated, users can edit it using the standard tools available in Microsoft Paint.

GPT Store Delayed: OpenAI has announced a delay in the launch of their GPT store, initially planned for release this month, to early next year. This postponement, communicated via email, is attributed to the unexpected developments we all know about that have preoccupied the team. In the meantime, OpenAI has promised upcoming updates to ChatGPT.

NVIDIA’s Jensen Huang and Elon Musk think AGI is coming soon: Recent discussions in the tech community have centered around predictions for the development of Artificial General Intelligence (AGI). Nvidia's CEO, Jensen Huang, at the NYT DealBook Summit, estimated that AGI could be achieved in about five years. Huang defines AGI as having intelligence comparable to a normal human. In contrast, Elon Musk is more optimistic (or actually pessimistic because he thinks AGI will kill us alll), suggesting AGI might arrive in just three years, with his definition focusing on surpassing the smartest human in any domain, including writing novels or inventing new technologies. These predictions highlight varying perspectives on the timeline and definition of AGI.

[Closing Thoughts]
What We Learned

We learned that the trend continues : huge announcements in the AI world and an increasing amount of breaches in the cyber world.

Like I write in the introduction , something’s got to give here. We can not continue like this.

It’s something to think about - but as I also said this is something to discuss in a bar late at night.