Scam Altman

Good morning,

We’re taking a bit of a breather this week at the Blacklynx Brief. Got some good reactions onlast week’s nuanced view. Where we studied the case against AI.

If there’s one thing you should remember from last week’s newsletter it’s that operating an LLM is “not cheap”. AI’s detractors are saying that it’s so unsustainable that we’ll see companies like OpenAI go bankrupt if they don’t do something quick.

It seems that these predictions are coming true: OpenAI just announced a new business model in which they sell you “agents” that will cost between 2000$ or 20.000$ per month. The idea is that this will replace a salaried human - 2000$ for a programmer - 20.000$ for a Phd-level employee.

I’ve always tried to keep personal opinion out of this, but I find this move reprehensible. It means that they are cordoning of AI capacity and create a distinction between the corporate world - giving them free reign to eliminate their employees and it keeps the real power away from regular users like you and I.

Just like AI’s biggest critics predicted - all the power will be in the hands of some happy few and who knows what will happen then. All this talk about doing it to the benefit of humanity seems cheap.

Looks like Elon Musk might be correct when calls the CEO of OpenAI ‘Scam Altman’.

At this point you have to really ponder if you’re still willing to pay for that ChatGPT subscription.

Lots to think about.

It’s a short one this week - enjoy the rest of the day and the weekend and see you next week where we’ll talk about “Vibe coding”.

Welcome to the Blacklynx Brief!

AI News

• OpenAI has launched GPT-4.5, a model focused on deeper world knowledge and emotional intelligence rather than math or scientific reasoning. While it hallucinates less and performs well on coding tasks like SWE-Lancer, it doesn’t represent a major leap over previous models, and its API pricing is significantly higher at $75/$150 per million tokens. With its steep cost and more subtle upgrades, GPT-4.5 feels like a refinement rather than a groundbreaking release.

• Tencent has unveiled Hunyuan Turbo S, an AI model designed for speed rather than deep reasoning, performing twice as fast while matching GPT-4o and DeepSeek V3 on benchmarks. The company has dramatically lowered costs and is preparing a separate T1 reasoning model for complex tasks, embracing a “fast vs. slow” AI strategy. As China’s AI race accelerates, Tencent’s latest move highlights how competition is shifting toward optimizing models for different thinking styles.

• Ideogram 2a delivers high-quality image generations in just 10 seconds, with a 2a Turbo version offering even faster results while cutting costs by 50% compared to Ideogram 2.0. The model excels in graphic design and text generation, creating detailed homepages, movie posters, and ad content with improved photorealism. As AI-generated design tools advance, 2a pushes creative automation to new levels, making professional-quality visuals more accessible.

• Apple’s AI-powered Siri overhaul has been delayed until at least 2027, with integration issues, low adoption of Apple Intelligence, and internal struggles slowing progress. As rivals like Alexa+ advance rapidly, Apple’s lag in AI assistants is becoming a growing competitive risk.

• Oculus co-founder Brendan Iribe’s startup Sesame unveiled a voice AI that mimics natural human speech with real-time context awareness and emotional adjustments. The company is also developing AI-powered glasses, signaling a shift toward more lifelike and responsive digital assistants.

• OpenAI confirmed plans to integrate Sora into ChatGPT and teased a faster Sora Turbo model alongside a new image generator. While Sora’s rollout has been slow, putting it inside ChatGPT could improve adoption, but competitors like Google’s Veo 2 are pushing the bar higher.

• T-Mobile’s parent company, Deutsche Telekom, is developing an AI Phone in partnership with Perplexity, making AI the core of the smartphone experience. The device will feature Perplexity Assistant on the lock screen and integrate AI partners like Google Cloud, ElevenLabs, and Picsart for real-time translation, content creation, and avatar generation. Set for release later this year under $1,000, this marks a shift toward AI-driven mobile interfaces and a major win for Perplexity in the AI space.

• Anthropic has closed a $3.5B Series E funding round, tripling its valuation to $61.5B just days after launching Claude 3.7 Sonnet with hybrid reasoning and coding capabilities. Led by Lightspeed Venture Partners with backing from Salesforce, Cisco, and Fidelity, the investment will support model development, AI safety research, and international expansion. With Amazon’s previous $8B stake and Claude models now powering Alexa+, Anthropic is cementing itself as a top AI player alongside OpenAI, Google, and xAI.

• Microsoft has introduced Dragon Copilot, a voice-activated AI assistant designed to streamline clinical documentation and reduce administrative burden for healthcare professionals. Merging Dragon Medical One and DAX Copilot, it automates clinical notes, referral letters, and workflow tasks, saving doctors an average of five minutes per patient encounter. Launching in the U.S. and Canada in May 2025, Dragon Copilot is part of the growing wave of AI tools reshaping the medical industry.

• Amazon is developing an advanced reasoning AI under its Nova brand, set for a June release, marking its biggest move yet to compete with OpenAI, Anthropic, and Google. The model will feature a hybrid reasoning system for both quick answers and deep problem-solving while aiming to undercut competitors on pricing. Despite its $8B stake in Anthropic, Amazon is making a clear push to establish itself as a major AI contender across multiple areas, including Alexa+ and AGI research.

• Cohere For AI has launched Aya Vision, a multimodal AI capable of vision-language reasoning in 23 languages, covering over half the world’s population. The 8B and 32B versions outperform much larger models, interpreting images, answering visual questions, and translating visual content across languages from Vietnamese to Arabic. Released under a CC non-commercial license, Aya Vision pushes AI accessibility further, breaking language barriers in visual understanding.

• OpenAI has introduced NextGenAI, a $50M academic consortium partnering with Harvard, MIT, Oxford, and 12 other institutions to fund AI research and education. The initiative provides grants, compute resources, and API access to advance AI-driven breakthroughs, from rare disease diagnostics to historical text preservation. With AI integration accelerating in academia, programs like NextGenAI and ChatGPT Edu signal a broader shift toward AI-powered research and learning.

• OpenAI is preparing to launch a suite of specialized AI agents priced between $2,000 and $20,000 per month, targeting business professionals, software developers, and Ph.D.-level researchers. Investor SoftBank has already committed $3B for 2025, with OpenAI projecting that agents will generate up to 25% of its long-term revenue. With prices rivaling senior employee salaries, this move signals a shift toward AI replacing high-level knowledge work in enterprise settings.

• Google has introduced AI Mode, a Search Labs experiment that turns traditional search into a conversational experience powered by Gemini 2.0. The feature gathers information from multiple sources simultaneously, allowing users to refine searches through follow-up questions, while AI Overviews has been upgraded for better performance on complex topics. As AI-driven search continues to evolve, Google faces growing competition from Perplexity, Grok, and ChatGPT, making this a crucial update to maintain dominance.

• Alibaba’s Qwen team has released QwQ-32B, a compact reasoning model using large-scale reinforcement learning to match or outperform bigger models like DeepSeek-R1 at a fraction of the cost. The model is 20x smaller than R1 yet delivers similar results, priced at just $0.20 per million tokens—90% cheaper than competitors. With its open-source release under Apache 2.0, Qwen continues to push the boundaries of efficient AI, bringing near-frontier intelligence closer to on-device use.

Quickfire News

• Pika Labs released Pika 2.2, featuring improved quality, 10-second 1080p video generations, and new transition and transformation capabilities.

• Meta is reportedly planning to launch a standalone Meta AI app in Q2, with potential paid subscription options similar to OpenAI's model.

• Figure is accelerating plans to bring humanoid robots into homes, beginning Alpha testing this year due to advancements from its recently unveiled Helix AI.

• Microsoft updated Copilot, adding a dedicated macOS app, support for uploading PDF and text files, and various UI improvements.

• Meta introduced Aria Gen 2 glasses, featuring advanced sensors, on-device AI processing, and an all-day battery for research in machine perception, contextual AI, and robotics.

• You Labs unveiled ARI, a research agent capable of analyzing up to 400 sources and generating professional reports with charts, citations, and visuals in under five minutes.

• DeepSeek revealed that its AI models theoretically generate 545% profit margins on inference costs, in stark contrast to U.S. rivals currently operating at a loss.

• Anthropic CEO Dario Amodei said the company is “reserving” Claude 4 models for major advancements and predicted AI will surpass the best human coders by 2026.

• SoftBank is reportedly seeking $16 billion in loans to fund its AI investment strategy, with Elon Musk commenting that CEO Masayoshi Son is “already over-leveraged.”

• Anthropic will participate in the Department of Energy’s “1,000 Scientist AI Jam,” where its Claude model will be evaluated for scientific research and national security applications.

• Samsung launched new $300 Galaxy A series phones, bringing AI features like Circle to Search and AI-powered photo editing to compete with Apple’s $599 iPhone 16e.

• Chinese smartphone giant Honor announced a $10 billion AI investment plan to transition into a global AI device ecosystem company.

• TSMC announced an additional $100 billion investment in the U.S., raising its total commitment to $165 billion for five new chip plants in Arizona.

• The latest version of Grok-3 moved to the No. 1 spot on the LM Arena leaderboard, overtaking GPT-4.5 Preview just hours after its debut.

• Stability AI partnered with Arm to bring Stable Audio Open to smartphones, achieving 30x faster on-device audio generation without requiring an internet connection.

• Google introduced Data Science Agent in its Colab coding environment, a tool that automates data analysis by generating complete, working notebooks.

• Podcastle launched Asyncflow v1.0, a text-to-speech AI model with 450+ voices and API access, featuring voice cloning from just seconds of audio input.

• Google announced that Project Astra’s live video and screen-sharing features will roll out to Gemini Advanced subscribers on Android this month.

• Google’s Pixel 10 will reportedly introduce Pixel Sense, an on-device AI assistant capable of processing data from 15+ Google apps to complete complex tasks.

• Tencent’s Yuanbao AI app overtook DeepSeek as the top downloaded iPhone app in China this week, following the release of its “fast-reasoning” Hunyuan Turbo model.

• ASLP Labs introduced DiffRhythm, an open-weights model that can generate complete 4-minute songs with vocals in just 10 seconds using lyrics and style prompts.

• Amazon created a dedicated agentic AI group within AWS, with CEO Matt Garman calling it a “potential multi-billion business” focused on task automation.

• Cortical Labs launched CL1, the world’s first commercial “Synthetic Biological Intelligence” system, integrating living human brain cells with silicon hardware.

• Cornell and Tel Aviv researchers unveiled ProtoSnap, an AI tool that deciphers 3,000-year-old cuneiform tablets by matching template characters to ancient texts.

• CoreWeave, ahead of its IPO, acquired AI developer platform Weights & Biases, planning to integrate it into its cloud infrastructure offerings.

• OpenAI CEO Sam Altman said GPT-4.5 will roll out to Plus users in phases “over a few days”, and hinted at a credit-based system for accessing advanced features like Sora.

• Digg is being revived by founder Kevin Rose and Reddit co-founder Alexis Ohanian, featuring AI-enhanced moderation and UX.

• OpenAI expanded access to GPT-4.5-Preview, rolling it out to all Plus users after initially launching for Pro and API developers last week.

• A federal judge denied Elon Musk’s request to block OpenAI's transition to a for-profit, but allowed other parts of his lawsuit to move forward.

• Andrew Barto and Richard Sutton won the 2024 Turing Award for their pioneering work in reinforcement learning, while cautioning against rapid AI deployment.

• Scale AI secured a multimillion-dollar U.S. DoD contract for Thunderforge, a program aimed at deploying AI agents for military planning and operations.

• Codeium launched Windsurf Wave 4, adding features like AI-powered previews, tab-to-import functionality, and suggested actions for developers.

• Luma Labs introduced three new features for Ray2, including Keyframes, Extend, and Loop, giving users more control over AI-generated videos.

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 !