Elon Derangement Syndrome

Good morning,

It’s becoming increasingly difficult and even impossible to catch up with AI News. I try to tune out anything political, but the worlds of technology and the world of politics are intersecting in some very interesting AND disturbing ways.

On the technology front—what a week again. The breakthroughs keep coming in hard and fast.

Microsoft has reportedly gained a breakthrough in quantum computing and after 17 years of research developed a chip with a new type of matter - that could be the start to scaling quantum computing. (But they’re still not able to fix Teams or Outlook search).

A second big story was the Arc Institute’s EVO2 project. Evo 2 is an advanced AI that analyzes vast amounts of DNA from across all life forms, enabling it to predict genetic impacts—like disease risks—and design custom DNA sequences for applications such as precision medicine or sustainable agriculture. It’s trained on 17 trillion DNA sequences found in nature and is able to write and sequence DNA. This is a big deal!

But the biggest and certainly the loudest story was the release of X.AIs Grok3.

I decided to feature Grok3 today.

But if you talk about Grok3 you have to talk about Elon Musk.

Elon Musk has become a very polarizing figure (as I said technology intersecting with politics - toeing a fine line between genius and downright batshit insane).

But we have to talk about him and x.AI

He’s so polarizing that I believe I’m going to lose some subscribers today.

If you already feel the itch, please bear with me and let me explain.

Two years ago, I dedicated one of this newsletter’s early editions to Elon Musk’s accomplishments in AI, and I took some serious flak for it. People seem to have a visceral reaction to Musk. The reaction to his ‘bad’ behavior is so emotional they cannot see the ‘good’ he’s been doing anymore. At that time I defended him, but I can’t do that anymore.

But he’s not evil incarnate.

After Trump Derangement Syndrome, there is now Elon Derangement Syndrome. Where your hatred for a person is so intense you cannot see anymore if and when this person does something positive.

When the crazy things a person does wipes away any positive counterpoint.

It’s going to be so much worse this time around. Because Elon Musk has polarizing the world in the extreme.

But when it comes to Musk, I believe the good still outweighs the bad. Although the case against him keeps piling up.

But you have to step back and look at what he’s doing for humanity in general. If you dedicate the time and listen to one of the many three+ hour interviews he has done and you can set aside some of your feelings, you will realize that he thinks in a really unique and interesting way. We need people like that.

Starlink for example is an extremely valuable product - it’s an amazing idea, wonderfully executed. It allows people that are living in a cabin in the Arctic access to the internet. But it’s also a powerful military weapon — which he is using to assert geopolitical power.

It’s the perfect example of the complexity of Elon Musk.

On top of that, Elon Musk has been caught in multiple lies lately, even going so far as to tell the world he’s the number one player in both Diablo IV and Path of Exile , two popular arpg’s (action role playing games). While in reality he just paid off people to play on his account 24/7.

He’s also not winning any father of the year awards.

His behavior indeed seems erratic and highly questionable lately: the infamous nazi salute, his cozying up to the far right AfD in Germany, his ongoing efforts with “DOGE”, him living in Mar-a-Lago and bringing one of his 20 kids into the White House for a bizarre press conference with Donald Trump. And I’m not saying that’s ok, but he shouldn’t be cancelled over this.

That would be like silencing Einstein before he was able to work out his relativity theory- therefore depriving humanity of a potential breakthrough.

Anyway, I read Elon Musk’s biography and just like with Steve Jobs’ biography by the same author my conclusion was more or less the same : both people are(were) giant assholes, are both sociopaths, and are a complete nightmare to work for.

But they’re both geniuses, and you cannot deny that both men have done more for humanity than most of their fervent critics combined (that includes you and me).

Hence why Musk is the richest man in the world (and perhaps the most hated simultaneously)

Elon Musk is going to get us to Mars. He’s going to fill our factories with humanoid-looking robots. He is going to make full self-driving a reality and save thousands upon thousands of yearly car deaths. And Elon Musk is probably going to win the race for AGI.

But he’s certainly not all talk and this week’s Grok3 release is proof of that.

Enter Grok3

A few years back Elon Musk founded X.AI with a distinct mission.

This is what he had to say about it : "The mission of xAI and Grok is to understand the universe. We want to answer the biggest questions: Where are the aliens? What’s the meaning of life? How does the universe end? To do that, we must rigorously pursue truth, even if that truth is politically incorrect or challenges what people want to believe."

This monday - X.AI debuted Grok3. It’s early days but it’s topping all leaderboards, surpassing Deepseek, OpenAI, Google, Meta, …

Also on chatbot arena it is destroying the competition:

I have to say, the early results are amazing. It’s the result of X.AI’s 100.000 piece graphics card cluster being trained for more than a month.

However it’s still early days and while it might be topping the leaderboards - it also seems to fail at some very basic questions. It’s also prone to lying (just like its spiritual father).

We’ll be looking deeper into this model as the weeks progress but for now it scores a perfect 100% on my own benchmark:

Welcome to the Blacklynx Brief!

AI News

• Anthropic is preparing to release a new AI model in the coming weeks that will combine standard language skills with advanced reasoning, allowing developers to adjust speed and processing power as needed. The model is said to excel at coding and large-scale programming tasks, with a sliding scale feature to control the level of reasoning for each query. After months of quiet, this launch could reestablish Anthropic as a top competitor as AI models increasingly focus on hybrid capabilities.

• YouTube is upgrading its Dream Screen feature with Veo 2, allowing creators to generate entire video clips and backgrounds from text prompts in Shorts. The tool can apply styles, camera effects, and realistic physics, with all AI content labeled using Google’s SynthID watermark. Rolling out first in the U.S., Canada, Australia, and New Zealand, this update brings cutting-edge AI video production into everyday content creation.

• Galileo Labs introduced a leaderboard evaluating 17 leading AI models on their ability to use tools, process long context, and handle complex tasks, with Gemini 2.0 Flash and GPT-4o topping the rankings. Open-source models like Mistral’s latest are catching up, while DeepSeek models were left out due to limited tool-use capabilities. As AI agents become a major focus, this kind of performance tracking is expected to guide businesses in choosing the right models for their needs.

• Meta is building AI, hardware, and software platforms for humanoid robots, aiming to power other manufacturers’ products rather than creating its own. A new team within Reality Labs, led by former Cruise CEO Marc Whitten, will develop core systems and has held talks with robotics firms like Unitree and Figure AI. Meta’s strategy mirrors Android’s role in smartphones, signaling a unique platform-first approach as tech giants push into robotics.

• Perplexity has introduced its own Deep Research tool, which compiles in-depth reports from hundreds of sources in minutes, outperforming some competitors like Gemini in testing. Unlike OpenAI’s $200/month version, Perplexity offers 5 free uses per day, with more for Pro subscribers, escalating competition over AI-powered research assistants.

• The NBA showcased new AI and robotics systems at its All-Star Technology Summit, including A.B.E. for passing and rebounding, M.I.M.I.C. for simulating plays, and K.I.T. for player wellness and motivation. Stars like Stephen Curry are already using the tech in training, reflecting the growing role of robots in professional sports. As teams chase every possible edge, AI-powered tools are set to become a standard part of elite athlete development.

• As mentioned above, xAI has unveiled Grok-3, a new AI model outperforming GPT-4o, Claude 3.5 Sonnet, and Gemini-2 Pro on key math, science, and coding benchmarks, with a smaller Grok-3 mini version also launching. Trained on 10 times more compute than Grok-2 using xAI’s massive Colossus supercomputer, the model has reasoner capabilities similar to OpenAI’s o3-mini. While Grok-3 puts xAI briefly at the top, rivals like OpenAI’s GPT-4.5 and GPT-5 are on the horizon, keeping competition fierce.

• French startup Mistral has launched Saba, a 24B parameter AI model trained on Arabic and South Indian languages like Tamil and Malayalam, built for culturally accurate content and conversational AI. It offers faster, cheaper performance compared to larger models and is available via API and for local deployment.

• The New York Times is allowing journalists to use AI for non-writing tasks like SEO, research, and article summaries while still banning AI-generated news content. The paper introduced Echo, an internal AI summarization tool, and approved platforms like GitHub Copilot, Google Vertex AI, and OpenAI’s API. Despite ongoing legal battles over AI copyright, major publishers, including Financial Times and Associated Press, are increasingly integrating AI into newsrooms to boost productivity.

• Google has introduced an AI-powered research assistant that uses six specialized agents working together to generate and validate scientific hypotheses in fields like medicine and genetics. Tested at Stanford and Imperial College, it identified new drug uses and gene transfer mechanisms within days, achieving over 80% accuracy on expert-level benchmarks. Google is rolling out the tool through a Trusted Tester Program, positioning AI as a future key partner in scientific breakthroughs.

• Microsoft researchers have developed Muse, an AI system capable of creating two-minute, playable game sequences from just one second of reference gameplay. Trained on seven years of data from the game Bleeding Edge, Muse generates realistic 3D environments and actions following proper game physics. Microsoft is open-sourcing the model, which could significantly speed up game development by cutting animation and design time from months to days.

• Arc Institute and Nvidia have released Evo 2, a genome AI model trained on 9 trillion DNA sequences from 128,000 species, enabling it to analyze entire bacterial genomes and human chromosomes at once. It achieved 90% accuracy in predicting cancer-causing mutations and successfully designed synthetic genomes, with the 40B-parameter system now freely available through Nvidia’s BioNeMo platform. Evo 2 marks a shift toward AI systems that can understand and design biological life on a large scale.

• Mira Murati has unveiled Thinking Machines Lab, a new AI research company aimed at building advanced models for science and programming while promoting human-AI collaboration. She’s assembled a top team, including OpenAI’s John Schulman and talent from DeepMind and Mistral, with a commitment to open science through regular releases of research, code, and model details.

• OpenAI has introduced SWE-Lancer, a benchmark assessing AI models' coding abilities using 1,400 real freelance tasks from Upwork, with theoretical earnings totaling $1 million. Claude 3.5 Sonnet topped the leaderboard, solving nearly half of the tasks and “earning” $400k, though all leading models struggled with the more complex work.

• Fiverr has introduced Fiverr Go, a set of AI tools enabling freelancers to train models on their work for $25/month and automate client tasks with a $29/month assistant. The platform is also giving top-performing freelancers company equity, aiming to help workers benefit from AI instead of being replaced by it.

Quickfire News

• Over a dozen major news publishers filed a lawsuit against Cohere, accusing the company of copyright infringement and trademark violations for using their content to train AI models and generating articles that mimicked their brands.

• Baidu announced plans to make its Ernie chatbot and advanced search tools freely available starting April 1, aiming to drive adoption amid increasing competition from DeepSeek.

• Apptronik raised $350 million in a Series A funding round, with plans to scale production of its Apollo humanoid robot and expand into healthcare and consumer markets.

• Elon Musk’s letter of intent to acquire OpenAI was disclosed, offering $97.4 billion in cash with a May 10 deadline and conditions requiring full access to company records.

• OpenAI launched an updated version of its GPT-4o model, improving performance in creative writing, coding, and instruction following.

• Robotics startup Figure AI is reportedly in talks for a $1.5 billion funding round, which would raise its valuation to $39.5 billion.

• Google rolled out memory upgrades to Gemini Advanced, enabling the model to recall and reference past conversations in future responses.

• Apple is reportedly planning to introduce Apple Intelligence features to its Vision Pro headset in April, including Writing Tools, Genmoji, and Image Playground.

• Ilya Sutskever’s startup SSI is reportedly in talks to raise over $1 billion, pushing its valuation beyond $30 billion just months after launch.

• Nous Research released DeepHermes-3, an 8B parameter open-source model with a toggle feature allowing users to balance reasoning quality and speed.

• OpenAI published a prompting guide for its o-series reasoning models, advising users to favor simpler, more direct prompts over complex instructions.

• SoftBank’s Arm is reportedly planning to develop its first in-house AI chip, with Meta expected to be an early customer, marking a shift from Arm’s usual licensing model.

• OpenAI CEO Sam Altman posted that GPT-4.5 testers have experienced "feel the AGI" moments, adding to speculation about an upcoming model release.

• Chinese AI startup DeepSeek suspended chatbot app downloads in South Korea after regulators raised concerns about data privacy practices.

• Perplexity open-sourced R1 1776, a retrained version of DeepSeek’s reasoning model that maintains performance while removing built-in censorship.

• Apple introduced the iPhone 16e, its most affordable Apple Intelligence-enabled device, starting at $600 and featuring the company’s first in-house 5G modem.

• Convergence AI launched Proxy 1.0, a free web agent capable of clicking, typing, and navigating websites to automate online tasks for users.

• Clone Robotics shared new footage of Protoclone, a bipedal android with a musculoskeletal structure and 500 sensors, designed to mimic human anatomy and movement.

• OpenAI CEO Sam Altman posted a poll on X asking what users would like open-sourced, with an "o3-mini" level model leading over a "phone-sized model."

• Elon Musk announced the launch of xAI’s Gaming Studio during a Grok-3 demo, with plans to develop AI-powered games.

• xAI revealed that a Voice Mode for Grok will go live in about a week, sharing a brief teaser at the end of its recent demo.

• HP acquired Humane’s AI software platform and team for $116 million, discontinuing the AI Pin hardware and planning to integrate AI features across its devices.

• Meta announced Llamacon, its first dedicated generative AI developer conference, scheduled for April 29.

• Google added new AI-powered features to Google Meet, including a scrollable caption history that lets users review up to 30 minutes of live and translated captions.

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !