Every week, get the latest curated high signals on news, insights, and ideas around technology, science, and business to help you become a better builder and thinker. Please fill out the survey so we can deliver the best content to your inbox. Thanks for reading!
I came across this study sent to me from friend and thought it would be interesting to share. I’ve always wondered what kind of background sound actually helps people, especially those who struggle to stay on particular tasks. In my case, I would usually just throw on a chill playlsit or some lofi beats on repeate and hope for the best, but it appears that hit-or-miss approach ignores how our brains crave just the right amount of stimulation, which varies person to person. Most research on this topic compares music or noise to silence, so we never really know which specific musical tweaks matter, or why the same track can boost one person and irritate another.
The researchers ran several expeirments where people did attention tasks while listening to different sounds: normal music, pink noise, and music with quick, repeating volume “pulses” (AM+music). The “AM” part refers to Amplitude Modulation. They found that the music with these fast pulses helped people focus better, but only when it was played first. Brain scans showed this pulsed music got the attention parts of the brain working harder, and brainwave tests showed peoples’s brains were syncing up with the music’s rhythm. When they tried different speeds of these pulses, they discovered that at a certain speed (~16 times/second), it was best for people who have a hard time focusing, like those with ADHD symptoms. On the visual below, just focus on the last item in the first column. You can see a spike as a response of the AM+music audio being induced into the subject versus the other methods.
The takeway here is: it isn’t actually about the mood or how “hype” the song feels. It’s the precise rate and depth of those hidden volume pulese which makes the difference.
Here’s this week’s scoop:
GPT-5: LLM progress could be slowing down
Is AI stealing our jobs?
India wants to build its own national AI model
The US might buy a piece of Intel
The surge in robots
Databricks is seeking to raise at $100 billion valuation
NASA is building an AI that can predict solar storms
🔥 Nuggets for the Road
Canva begins share sale at $42 billion valuation [LINK]
A new way to make quadratic equations easy (finally!) — overlooked for 4,000 years, there’s now a new way to solve them! [MIT]
Renewables are becoming much more affordable — more than 90% of last year’s global renewable energy capacity was cheaper than a fossil fuel alternative [LINK]
Ranking all the Chinese open model builders — the irony is that the open source movement is being driven much more by China than the US, something to think about… [LINK]
📡 The Signal
GPT-5: LLM progress could be slowing down
The general problem in AI is that each new model is hyped up as a massive leap forward (in some cases, it is!), but most of the time, the upgrades don’t live up to the anticipation. The playbook is that these hyperscalers would throw more money and computing power at the product, but that approach is starting to hit a wall. GPT-5 (released last week) exemplifies this trend. It’s definitely shown some improvements, i.e., being better at coding, writing, and even being your friendly doctor, but it’s not the radical step change we were all expecting. Even the benchmarks show it’s only as good as human experts about half the time. Don’t get me wrong, I’m all about incremental changes and progress, but we’re now slowly seeing these companies misalign the reality of these models with the overhyped promises they’re making. I’m thinking that the whole industry might need a new way of building AI. It’s a question of when rather than if, but I don’t think we’ll be reaching AGI this year. Your jobs are safe. [LINK]
AI vs Jobs: Not as bad as we thought
Everyone’s talking about how AI is supposedly killing jobs, and most people just assume AI is the ultimate culprit, but that couldn’t be further from the truth. Typically, you would usually look at unemployment rates as a barometer for job displacement, but other layers of the job market need exploring to confirm if AI is really having an effect on labor. The researchers divided jobs into five groups based on how much AI could perform those tasks, using not one, but five different methods to measure exposure. They also looked at things like who’s quitting work, switching careers, or how companies might be shuffling tasks around. The other thing is that they didn’t want to focus on just workers, but also what’s happening at the company and industry level, too. The major finding is that across all these areas and measures, there wasn’t really a big sign that AI is ruining the job market. If anything, people most exposed to AI tend to have better pay, more education, and lower unemployment than those less exposed. I don’t think there’s a mass exodus from the workforce, and people aren’t rushing to switch to “safer” jobs. Even in industries using plenty of AI, employment is mostly steady or even growing.
The visual below shows the changes in the unemployment rate, with (1) being the least exposed to AI, showing the biggest rise, and (4) being the most exposed to AI, showing the smallest rise. [LINK]
India wants to build its own national AI model
India is the next superpower that’s thinking about building their own foundation model, but there are a few hurdles. The biggest problem is the languages, since they have 22 official ones and hundreds of dialects, which makes building AI models tricky. Most countries, like the US and China, train their AI primarily using English data, but in India, there’s not a lot of high-quality online content in their native languages. Usually, teams try to solve this by cramming in whatever data they can find, but that doesn’t work well because the models end up misunderstanding Indian languages (nuances, context, etc.), so the results end up pretty weak. The other facet to this is that historically, India’s technology sector has focused more on IT and software services, especially in support of foreign companies requiring more affordable labor. So when it comes to creating frontier tech, it’s been a bit lackluster — although this is slowly changing. The software industry has spent way less on R&D compared to other countries, so their AI models haven’t been able to compete globally. Now, with the advent of DeepSeek in China, the Indian government is realizing it’s falling behind. The government is now becoming more active in helping foster tech innovation across AI. There’s a push to make more data available, fund deep research, and support startups to really develop homegrown AI solutions. [MIT]
The US might buy a piece of Intel
As of late, Intel has been struggling to get its huge factory off the ground. The typical route in these cases is to maybe cut costs, get private investors, or ask for government incentives. But this time, it’s unique because the stakes are high in making the US stay ahead of the chip game against countries like China. The Trump administration is now considering buying a stake in the company. It’s unusual since it usually keeps its distance from owning companies, for the sake of helping Intel get the factory up and running. This is not a bailout plan, as was the case back when the big auto manufacturers needed support. Instead, the government is actually owning a piece of Intel, which could change how the company works and how much say the government has in the tech industry. It’s a pretty brazen move on behalf of the US, but it simply goes to show what lengths they’re willing to go to outcompete other countries on this front. [LINK]
The surge in robots
Robots are making a huge comeback. A few weeks ago, I wrote about the rise of humanoids and their amazing progress, but there are still huge developments in industrial robots that are the workhorses behind the things you and I take for granted. Robotics is moving so fast that if industries fail to adopt quickly, it’s fair to say they will probably end up stuck, less efficient, and unable to keep up with competitors who are doubling down on tech. Now, some key sectors (like automotive, electronics, etc. ) aren’t holding back and buying more robotic infrastructure. I think the huge difference now is that companies aren’t really chasing efficiency gains anymore. Automation is now table stakes and their way of staying flexible and competitive. Non-automotive industries actually ordered more robots in the second quarter, showing that automation is spreading. There’s also a big spike in “cobots”, which are collaborative robots, which are designed to work safely alongside people, which is handy in places that are short on space for workers. Cobots made up almost 25% of all robots sold in Q2. [LINK]
Databricks is seeking to raise at $100 billion valuation
The company is aiming to raise fresh capital. ICYMI, Databricks is a company that builds software to help businesses organize, analyze, and use their data, especially for AI and analytics. You can imagine a big retail company with huge amounts of sales data from its stores and online. They can use Databricks by pulling all that data together, analyzing it, and building AI models that predict what products customers will want next month. So, for e.g., Databricks can help the retailer spot patterns, like which items sell more during certain seasons, or how online ads affect in-store purchases. Right now, companies have tons of data, but turning all that information into something useful isn’t as easy as it sounds. Traditionally, these businesses subscribe to different tools or rely on old databases that don’t really work well with AI, so you get really clunky systems. With GenAI becoming much more mature, the holy grail for enterprise is seeing how AI can be trained on a company’s data, which is why this raise is so crucial. Also, when it comes to raising rounds, this is one instance where additional capital is really valuable. Databricks doesn’t really need this money for day-to-day stuff because they’re already cashed up from past rounds — they generated $3.7B in annualized revenue this year! They’re raising cash simply because they’re going after something much bigger: making AI work better for enterprise and businesses. This is such a king move from Databricks, because they positioned themselves strategically to ride the AI wave at the right moment. [DATABRICKS, TECHCRUNCH]
NASA is building an AI that can predict solar storms
Surya (“Sun” in Sanskrit) is a new AI foundation model that was co-created between NASA and IBM to predict solar storms. Basically, solar storms are explosions from the sun that can mess with pretty much all electronics, from satellites all the way down to ground-based equipment on Earth like power grids. The problem is, while scientists can usually spot when the sun is going to create these storms, figuring out exactly when and how strong it’ll be is more challenging. Most of the time, predictions don’t really give us much warning. Surya was designed to solve this. They trained it on a massive data set of sun photos (over 250 Terabytes!) to see if it could spot patterns that humans miss. The plan is that Surya should be able to predict incoming solar flares two hours before they hit, which is about double the warning we usually get, and make a big difference for anyone relying on satellites. BUT, Surya isn’t just about predicting solar flares, but is designed to learn all sorts of things about the sun.
To give you more context, 15 years ago, NASA launched the Solar Dynamic Observatory (SDO) satellite. You can check out the website here. When it was launched, the AI boom had not taken off, so the collected images were not as valuable as they are today. Now with Surya, they’re able to harness AI to really process and understand these solar images in greater detail. [MIT, IBM]
Below is an image of how solar flares affect communications equipment and critical infrastructure here on Earth.
If you liked this, here are a few more popular posts:
To build the best newsletter for you all, I’d love to get some feedback. If you have some time, please fill out the following survey.
🏆 Alpha Bytes Premium
If you like our content, you can explore our Premium option. You can read more about it here. This includes (for now):
✅ Full access to the weekly newsletter ✅ Full access to podcasts + curated transcripts in Q&A format—this has become widely popular. ✅ Full access to all deep dives, essays, and tactical field guides ✅ Early access to new features and experimental content
🤩 Deals, Discounts & Affiliates
You can find carefully selected offers and deals here.
