Thomas Frey CSP

Mac users have a particular frustration not shared with Windows users: The inexplicable proliferation of dozens of "Noto [name]" fonts in their font menus, which started a few OS's ago. You can't uninstall or deactivate them (they're locked Apple system fonts), which means they infest your font menus in apps from #Adobe, #Microsoft, and other developers, despite font management programs you may try. In this short video I show a quick and dirty (okay not dirty) way to banish them from Adobe font menus—and no, I'm NOT talking about tediously Favoriting other fonts: https://lnkd.in/dFStmVVJ #noto #fonts #mac #apple

77

15 Comments

Power. It's a hell of a drug. Yesterday, I published an article on using Roblox as a model for work and culture design. The topic wasn't power, but a member asked an excellent question about it. Ethel Mendez asked "Why do you think it's so hard for some orgs and managers to give up a little power or control?" Good question. Complicated answer. 1. It all starts with everyone's favorite neurotransmitter, dopamine. Feeling in control and having power over another produces large amounts of dopamine. Dopamine is addictive. It's something we need to feel like we're thriving. The brain basically says, "Hey! Whatever you just did. Do it again, please!" In other words, even a little bit of power is pretty addictive. 2. Something interesting happens as a person's influence begins to expand beyond themself. It creates what psychologists call "approach orientation." It's a reduction in your overall social inhibition. You're no longer as concerned with what others think and feel because deep down your brain is constantly sending you signals that your judgement is superior due to your higher power status. It's so automatic that it bypasses our higher level thinking that would help us challenge the notion. 3. At the same time that we are receiving dose after does of dopamine and expanding our approach orientation, we try to protect ourselves from the pain that comes with loss of power. Three big things that creep in are: loss aversion, illusion of indispensability, and identity formation. I posted on status threat and loss aversion last week. For some, the loss of power is as painful as a punch in the face. To avoid that pain, leaders choose behaviors that protect their status as leader. On the more benign side it could be things like always having to have the last word. On the more extreme side, we're talking about things like career sabotage. The most common expression? Micromanagement. Illusion of indispensability is exactly what it sounds like. You are convinced that only you can accomplish what needs to be done. Then there's the fusion of your identity with the title of "leader." What happens over time is the title of "leader" becomes part of your identity. Without it, you begin feeling like you are missing a part of yourself. To reclaim that sense of identity, people will go to great lengths. It also, helps explain why some leaders have such a difficult time retiring. I had a leader many, many years ago. He was 71 and still saying he couldn't imagine retirement. He loved what he did too much to leave it. Well, that, and the fear of losing his status. He was candid in admitting that his title was who he was, and he really didn't know who he was if not that. Happy ending. He found another identity in his community. So, that's a very quick overview. It's obviously more complicated. But knowing the basics should help make sure you don't accidently become our good buddy Emperor Palpatine, who defines himself only by how much power he has.

75

8 Comments

The Creative Century is just getting started. Here’s another sign. NYC art schools are seeing record enrollment. Parsons reports a 64% increase in applications since 2016. Fine arts applicants at New York City public schools are up too. We’re now 25 years into the 21st century. When we see the forces that shape this century, we see the obvious ones — the internet, phones, AI. But there’s one that’s not quite as clear: creativity. As much as any other force, creativity shapes the world as we know it today. This surge in interest in art schools is happening amid historically low post-graduation employment rates and widespread AI fears. Students tell reporters: "Nothing is guaranteed anyway, so study what matters to you." We’re at the beginning of a mass explosion of expression. More kids want to be creators than astronauts — and they actually all can do it. The 21st century will be culturally, politically, and economically dominated by self-expression and creative output. The opportunity here is building creative infrastructure that serves this desire for meaningful collective work and ownership. Artist Corporations — a new legal structure I introduced at TED earlier this year — would give creative people the same economic tools available to other industries: shared ownership, access to capital, and the ability to build equity together, rather than compete alone. I hope many class of 2029 graduates form Artist Corporations.

91

11 Comments

Ever thought a painting could be a medical record? Researchers are now using #AI and X-rays to scan Old Master paintings — not just to find hidden sketches, but to reveal how pigments decay like living tissue. Even crazier: the patterns resemble medical scans of bones, lungs, and skin. Some labs now use paintings as models to test preservation techniques — techniques that later help in medical imaging. In other words, art is not just reflecting humanity… it’s teaching us how to heal it. Think about that: a portrait painted 400 years ago, holding clues that advance modern medicine. It reminds us that art isn’t only about beauty. It’s a living archive of chemistry, physics, and time itself. And if we look closely enough, the canvas can become a classroom — for science, for healing, for the future. 💭 If a painting can hold centuries of hidden stories, what else could it teach us about our own bodies? #ArtAndScience #Neuroaesthetics #InnovationInMuseums #SlowArt #CulturalData

175

10 Comments

"Whitney Museum of American Art curator explains the history of art versus digital tech. AI generated images are now seeping into advertising, social media, entertainment, and more, thanks to models like Midjourney and DALL-E. But creating visual art with AI actually dates back decades. Christiane Paul curates digital art at the Whitney Museum of American Art, in New York City. Last year, Paul curated an exhibit on British artist Harold Cohen and his computer program AARON, the first AI program for art creation. Unlike today’s statistical models, AARON was created in the 1970s as an expert system, emulating the decision-making of a human artist." https://lnkd.in/eVS3r9eW

40

Meta just dropped $16 million on a new audio lab in Cambridge. Can this change how we hear the world? Imagine a place where audio engineers can pretty much use magic to make sound feel like it’s coming from anywhere – not just left or right, but all around you. That’s what Meta’s new audio lab in the UK’s Ox-Cam corridor is built for. Some highlights from this new playground for audio geeks (like me): - Ultra-quiet test spaces (think: anechoic chambers big enough to fit a car) - One of the world’s largest reverberation rooms with 101 panels that can mimic anything from a phone box to a cathedral - Real home-like rooms with sensors to test wearables in real life - Motion tracking zones where AI learns how we move and listen Meta wants its smart glasses (Ray-Ban Meta, Oakley Meta) to adapt to your world. Loud train? Busy street? The tech will focus on what you want to hear and cut out the rest. That’s not just a nice-to-have – it’s a game changer for anyone who’s struggled to follow a call or enjoy a podcast in a noisy place. Why does this matter? Because the future of audio is immersive, intelligent, and personal. With this lab, Meta is betting big on the UK’s engineering talent and the power of AI to make audio smarter. The goal: sound that feels natural, connects us better, and includes everyone. As someone who lives and breathes 3D audio, I’m excited to see what breakthroughs will come from this lab. It just shows the importance of sound and that big players like Meta have understood this! What do you think about it? 🎧

756

42 Comments

WAVS.com, a popular sample library for music producers, made AI music headlines last week with this new feature. 💫 I'd compare the WAVS collection to Splice. They're the only companies in this space who meet my standards for music quality. 🔎 AI similar search: Upload audio to find similar samples on WAVS The new AI audio search tool combs through their 117,000+ files and returns the ones you're looking for in a matter of seconds. It's accessible on the website or with their DAW plugin. I'm personally stoked about the VST update as a Logic Pro user, because now I'm able to... ➤ Drag any audio or MIDI clip from my audio timeline directly into the WAVS plugin. Splice doesn't support this option with Logic. ➤ Retain the sound design of a MIDI file without bouncing it. Just drag the MIDI in and WAVS renders it automatically before searching. 😅 You mean I don't have to click through thousands of samples anymore? There's a running joke on instagram about what people *think* we're doing in the studio versus what we're *actually* doing... Too many producers spend hours gazing into the abyss, clicking through hundreds of snares trying to find the perfect one-shot. 🎸 When it comes to timbre, audio similarity works better than text filters Text filters are fine for broad categories like genre and instrument, but they're not effective for retrieving a specific tone quality. Timbre is usually described with words borrowed from other senses (warm, bright, dark). These qualities are hard to quantify and can be highly subjective. Audio similarity bypasses the language problem altogether, helping users find the music they're looking for. 💡 The obvious use case for audio search Something tells me that people are going to run trending songs through a free AI stem splitter like Sesh and ask WAVS to locate similar samples. I spoke to Daniel M., founder at WAVS, who confirmed that they don't train AI models on uploaded material or sell it to third parties. That means you can safely upload your own music, without the risk of having it gobbled up by some shady, involuntary opt-in down the road. 🤝 Artist fairness: The secret to WAVS' high quality music? The samples on Splice are all royalty free. That's convenient for musicians who use them, but it's not a great deal for people who made them. WAVS.com takes a different approach. Sample creators can choose from one of three tiered royalty models and display them transparently during search. As a result, people seem to be saving their best material for WAVS. There's an opportunity to make serious money if one of those artists writes a hit with your sample. Royalties only kick in if a song reaches 1M streams or gets released by a major label, so most small creators will never have to worry about managing a revenue share. 🩵 Nice to see AI music features that serve musicians, keep it up WAVS.com!

105

30 Comments

It's one thing to speculate whether AI poses an existential threat to humanity. But I wanted to know if Eliezer Yudkowsky and Nathan Soares, authors of "If Anyone Builds It Everyone Will Die" really thought THEY would be murdered by AI. Their answers: Yeah and Yup. Then I asked them how it would happen....https://lnkd.in/e6igxdNr

42

7 Comments

Client: "We think this is AI." Me, knowing it definitely wasn't: "How did someone come to this conclusion?" Client: "There were some words that AI likes to use, and many of the paragraphs were of similar length." Me: "Is that it?" Client: "Can you prove it wasn't AI?" Me: 🤦‍♂️ Folks, most of us have had this happen (especially with topics where there's very little room for creativity or differentiation). Blue Seven Content is not immune. With the volume of human-written work that we provide clients each week, I'm not surprised if, sometimes, something "looks like AI." The problem is that, without providing document histories for each piece to prove something isn't AI, there's no definitive proof to give. If someone has AI in their head, that's what they'll see. You have to trust the writers you work with. You should set clear expectations for what you expect. I'm not a fan of micromanaging writers to the point where I need to see each word they typed. They know what's expected of them. They know what the proper and improper use of AI is.

170

122 Comments

This great work co-led by Idan Shenfeld and Jyo Pari shows that online RL finetuning leads to less forgetting compared to standard SFT; and they find along the way that the RL solver inherently leads to a solution with a small reverse KL divergence! I'll try to discuss the significance of the result! When we do RL with a binary 0/1 reward, then reward maximization RL has many solutions all with a reward of 1 (assuming at least one such outcome exists). One can choose any distribution on the set of outcomes {y} that have a reward of r(x,y) = 1, and the expected reward E_π(.|x) {r(x, y)} would be 1. Let's call P* the set of all {π} with expected reward of 1. The authors prove that when doing online RL, you'd converge to the solution with the minimum reverse KL divergence to the reference model: min_{π ∈ P*} KL(π||π_0). Why is this important? We know that RL with reverse KL regularizer is equivalent to multi-tasking between the tasks encoded in the SFT checkpoint and the RL task. The multi-tasking helps you retain the capabilities encoded in the SFT checkpoint (Appendix C of https:// arxiv.org/abs/2412.19792). The result in this paper takes this a huge step further by showing that even if you don't use the reverse KL regularizer at all, online RL leads you to a solution with an inherent reverse KL regularization that helps achieve the same multi-tasking effect and retain the SFT checkpoint capabilities.

543

9 Comments

Beyond the Rhetoric of Theft: A Rebuttal to “The AI Revolution Is a Heist” Luke Savage’s article frames the AI revolution as a grand theft—an aggressive expropriation of cultural labor masked as innovation. While his critique highlights real concerns about power concentration and the exploitation of content, it collapses a nuanced debate into a single moral claim: that AI is nothing more than parasitic aggregation. This view, while rhetorically compelling, fails to engage with the complexities of fair use, creative transformation, and the evolving notion of contribution in the digital age. Savage overlooks the fact that all cultural production builds on existing material. Writers absorb libraries, musicians borrow riffs, and painters reinterpret visual traditions. AI’s training process, while industrial in scale, is not inherently theft—it is closer to a form of algorithmic “reading” and recombination, akin to how a student digests books before producing original thought. The key question is not whether AI should access cultural archives, but how it can do so within frameworks of fair use and just contribution. These principles—ensuring that original creators are recognized and, where appropriate, compensated—are not alien to technology. They can and must be integrated into AI governance, much like royalties or intellectual property rights evolved with past creative revolutions. By ignoring this possibility, Savage’s narrative risks fostering a sterile opposition between human creativity and machine systems. The reality is more complex: artists, educators, and thinkers are already using AI to expand their practices, testing its boundaries, and finding ways to inject intentionality and meaning into its outputs. These efforts should be seen not as capitulations to corporate power, but as acts of reclaiming agency over tools that will shape the future of cultural production. Yes, we must challenge the monopolistic tendencies of companies like Google and OpenAI, which currently operate in opaque ways and often fail to return value to the creators whose work underpins their models. But this requires structural proposals—new licensing models, transparent data sets, collective bargaining by creators—not just laments about a lost cultural purity. Without acknowledging the potential of shared stewardship and fair redistribution of value, Savage’s “heist” metaphor risks becoming another form of defeatism, seeing only predation where there is also the possibility of transformation. https://shorturl.at/8D1xk

30

12 Comments

artnet has launched their annual Intelligence report. I must assume that the report is based on their own sales data, ignoring market data on the blockchain. This is the reason why a lot of digital artists do not appear on the list of the most bankable artists of our time. Cryptopunks sales volume over the past 4 weeks is 33 mio. $ and 395 mio. $ over the past 12 months. Since the inception of Cryptopunks in summer 2017 almost 4 billion sales volume. Here are the numbers of the Artnet market report (first half of 2024): Pablo Picasso: 120 mio. $ René Magritte: 103 mio. $ Jean-Michel Basquiat: 101 mio. $ Roy Lichtenstein: 80 mio. $ Andy Warhol: 76 mio. $ The art market is generally weakening somewhat, which has also been reflected in the gallery closures of recent months. Some are citing structural market changes, which are due, among other things, to changing consumer behavior. In my opinion, we will see a rise of the digital art market in the coming years, a market that has yet to be canonized in art history. It cannot be ignored that the digitalization of recent decades has not only led to fundamental changes in our society, but has also had a significant impact on art. Blockchain technology and artificial intelligence, in particular, have played a key role. https://lnkd.in/ercUSe2a

45

7 Comments