Share

Yep it's going to discriminate real quick (+ it already does) + Big earnings + That Elon interview and more china inside
 β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ β€Œ

Colorintech Weekly - 307

(View this version on the web)

Hey


A big edition so check it out!


Check out the AI Podcast version of this newsletter or the Video version on our Socials


This newsletter is free, but if you do want to get us a summer ice cream🍦 as a thank you for 300+ editions grab us one here🎁

Oh and if you missed an edition, you can find it here or this platform, here

πŸ—žοΈDiversity and inclusion newsπŸ—žοΈ

πŸ€– AI Learns Bias Faster Than Humans


Turns out AI is more biased when it gets a chance than human


TL;DR πŸ‘€We already knew AI could inherit human bias. Turns out it can do something arguably more worrying: invent its own. New research found leading AI models developed stronger hiring biases than humans after just a handful of hiring decisions, suggesting that giving AI more memory and autonomy could create entirely new forms of discrimination. 🎭


What happened? πŸ“°

Researchers from Princeton University and the University of Chicago put ChatGPT, Claude, Gemini and other leading models through a simulated hiring exercise. All candidates were equally qualified, but after seeing just a few random successes and failures, the models began assigning entire fictional ethnic groups to particular jobs. πŸ“„

The reasoning models performed worst. OpenAI's o3 came close to the maximum possible score for stereotyping, showing around 65% more bias than human participants in the original psychology experiment the study was based on. πŸ“Š

The researchers say this isn't simply AI repeating biases from its training data. Instead, the models are optimised to spot patterns quickly. That makes them brilliant at maths and codingbut also surprisingly eager to generalise from tiny amounts of information when making decisions about people. 🧩


🧠 So what?

The industry has spent years talking about biased training data. This research suggests the next challenge is even trickier: AI systems may create new biases through experience, even when they start with perfectly balanced information. ⚠️

That's particularly relevant as companies race towards AI agents with long-term memory. Every interaction, correction and hiring outcome becomes another data point the model can use to build assumptions about people. The more autonomous these systems become, the more careful we'll need to be about what they're actually learning. 🧠

There was one encouraging finding. Simply telling the models to "be fair" barely changed anything. But rewarding them for diversity and giving them richer, relevant information about individuals significantly reduced bias. In other words, fairness isn't a prompt problem it's a systems design problem. 🎯

πŸ“š Read more

πŸ”— MIT Technology Review: AI is more likely than humans to form biases when hiring
https://www.technologyreview.com/2026/07/20/1140655/ai-biases-hiring-humans/

🧠Things that make you go hmmm🧠

πŸ”“Open or closed take your pickπŸ”“


TL;DR πŸ‘€

The AI industry's biggest battle is no longer OpenAI vs Anthropic. It's open vs closed. This week, Nvidia, Microsoft, Meta and 20+ tech companies urged governments not to restrict open-weight AI models, while China doubled down on making its best models freely available. The fight over who controls AI is becoming as important as who builds the best models. πŸ€–


What happened? πŸ“°

A coalition including Nvidia, Microsoft, Meta and Palantir published an open letter warning governments against placing "premature restrictions" on open-weight AI models, arguing that doing so would reduce competition and push innovation overseas. Notably absent? OpenAI and Anthropicβ€”the two companies whose business models depend on proprietary frontier models. πŸ“œ

The debate has intensified following the release of Chinese models like Kimi K3 and Qwen3.8, which are rapidly approaching the performance of the best American models while being available for developers to download, modify and run themselves. China's strategy is increasingly clear: make world-class AI widely accessible and let businesses build on top of it. 🌏

As Stratechery's Ben Thompson argues, the conversation is increasingly shifting away from who has the smartest model towards who can deliver intelligence at the lowest cost. If AI becomes a commodity, the winners won't necessarily be those with the best model they'll be those with the best economics and ecosystem. πŸ’°


🧠 So what?

This isn't really a technical debate. It's an economic one. πŸ€”

OpenAI and Anthropic want AI to look like AWS; a service you rent forever. China wants AI to look more like Linux something everyone can build on. Those are two fundamentally different visions for the future of computing. βš–οΈ


The irony is striking. While parts of Washington are debating restrictions on Chinese AI, American tech giants like Microsoft and Nvidia are effectively telling policymakers that limiting open models could hand China the advantage anyway. πŸ‡ΊπŸ‡ΈπŸ‡¨πŸ‡³

And here's the bigger picture. AI is no longer just about building the smartest chatbot. It's becoming a contest over standards, ecosystems and geopolitical influence. Whoever persuades the rest of the world to build on their AI stack could shape the next generation of software in much the same way Android, Windows and the internet itself did. 🌍

Meanwhile, back in the UK, our newly reorganised technology civil servants are still having coffee chats and getting to know their new colleagues. β˜•πŸ‡¬πŸ‡§

πŸ“š Read more

πŸ”— CNBC: Nvidia, Microsoft and Meta warn against restrictions on open-weight AI models
https://www.cnbc.com/2026/07/24/nvidia-microsoft-meta-open-weight-ai-models.html

πŸ”— Stratechery: Who's Afraid of Chinese Models?
https://stratechery.com/2026/whos-afraid-of-chinese-models/

models

πŸ€¦πŸΎβ€β™‚οΈThat elon interviewπŸ€¦πŸΎβ€β™‚οΈ


So Elon musk sat down with the Editor of The Economist Zanny Minton Beddoes last week for an interview. It was a lesson of incredible mansplaining but there were some things that we took from it 


1. Musk thinks superintelligence is five years away

Musk predicts that AI could become more intelligent than all of humanity combined within around five years. Within a decade, he believes there may be almost nothing humans can do better than machines.πŸ€”

Even more extraordinary was the lack of depth around what happens next. Musk’s position appears to have shifted from warning that AI could destroy humanity to accepting that the momentum is unstoppable and deciding we should β€œenjoy the ride”.

Perhaps we are old-fashioned, but β€œhopefully the superintelligence likes us” feels less like a safety plan and more like the final line in a film immediately before everything goes wrong. πŸ€–


2. His proposed AI regulator is… the AI companies themselves

Musk suggested that the leading AI laboratories should regularly inspect one another’s models, share early access and raise the alarm when a competitor creates something dangerous. Governments could then intervene where necessaryπŸ‘€

There is some logic to competitors scrutinising each other’s work. But it is still difficult to escape the basic problem: the same handful of companies racing to build superintelligence would largely decide whether their own race is safe.

It is the corporate equivalent of asking the foxes to establish a weekly working group on henhouse security. 🦊


His comments about Muslims and Europe were genuinely alarming

Musk argued that Europe is importing large numbers of people with beliefs β€œantithetical” to Western values and suggested that Britain could eventually face civil war as a result.

These were sweeping claims about millions of Muslims, made by someone who admitted he had not visited Britain for several years. The interviewer repeatedly challenged the dystopian version of the UK that Musk promotes to his audience, but Musk simply insisted that he was right.

There are legitimate conversations to be had about immigration, integration and social cohesion. But treating Muslims as a single cultural threat is not a serious analysis of any of them. It is the kind of framing that turns complicated policy questions into suspicion of an entire group of people.

The fact that it comes from someone with extraordinary reach and influence makes it more concerning, not less. 🌍


His β€œzero people died” claim on USAID showed astonishingly little depth

When challenged about the sudden removal of US aid funding, Musk repeatedly insisted that precisely zero (Yes you read that right) people had died as a result.

Not that the number was uncertain. Not that the evidence was contested. Zero.😯

He then suggested that organisations such as the Gates Foundation could simply step in and replace lost government funding. This seemed to ignore the fact that Bill Gates is already giving away vast amounts of his wealth and that Musk himself could presumably intervene if private billionaires are now expected to maintain global healthcare systems.

More importantly, international health programmes are not subscriptions that can be paused while everyone searches for a new payment method. When funding for medicines, clinics and staff disappears overnight, people do not politely wait for philanthropy to reorganise itself.

Musk may genuinely believe the number is zero. That would arguably make the answer even more extraordinary. πŸ’Έ


Musk appears to think followers are the same thing as fans

When challenged over how disliked he has become, Musk pointed to his enormous number of followers on X as evidence that more people like him than dislike him.

There are several problems with that argument.

Following someone does not mean liking them. People follow many politicians, journalists and public figures because they are newsworthy, controversial or powerful. Some may simply be waiting to see how low Musks stoop next.


We've seen reports where X has also previously made users follow Musk automatically, has long struggled with bots, and is ultimately a media platform itself. Musk criticises β€œthe media” as though X is somehow separate from it, while using X to shape the news, influence politics and broadcast his views to hundreds of millions of people.😩

He may be closer to Tom from Myspace than he realises: being placed in everyone’s network is not necessarily evidence that everyone wants likes you. πŸ“±


Sorry its paywalled but you can see the interview here or its been clipped up lots of places 


🧠The wider takeaway

The most striking part of the interview was not simply Musk’s confidence in the technology. It was the confidence with which he spoke about societies, cultures, international development and public opinion often with remarkably little evidence or curiosity.

This is someone who owns critical infrastructure, controls one of the world’s largest communications platforms and is helping to build the systems he believes may soon surpass humanity.

It would be comforting to hear more humility from the person telling us that human control may shortly become obsolete. πŸ‘€

πŸ’° Alphabet's AI Spending Spree Just Got Even Bigger


TL;DR πŸ‘€ Alphabet just posted another big quarter, with revenue up 24% and cloud sales surging 82%. But instead of celebrating, investors focused on one number: Google is now planning to spend Β£205 billion on AI infrastructure this year. The AI arms race just got even more expensive. πŸš€


What happened? πŸ“°

Google's parent company delivered another big set of earnings. Revenue climbed to $119.8 billion, driven by continued strength in Search, YouTube and subscriptions, while Google Cloud grew huge 82% to $24.8 billion. Its AI order backlog also reached $514 billion, suggesting demand isn't slowing down anytime soon. πŸ“ˆ


The surprise came elsewhere. Alphabet increased its planned AI infrastructure spending for 2026 from around $190 billion to $205 billion, with CFO Anat Ashkenazi signalling spending will rise again in 2027. Despite the strong results, shares fell sharply as investors questioned whether anyone not just Google is actually generating enough returns to justify these extraordinary levels of investment. πŸ’Έ

It wasn't just Alphabet. The announcement knocked confidence across Big Tech, with AI-related stocks selling off as investors started asking whether the race to build ever-larger AI data centres is becoming a very expensive game of one-upmanship. πŸ“‰


🧠 So what?

This is quickly becoming the defining question of the AI era: how much is too much? πŸ€”


For the past two years, investors have happily rewarded companies for announcing bigger AI investments. Now they're asking a different question: where's the payoff? Alphabet is spending more than ever, Tesla is pouring billions into robotics and autonomous driving, Microsoft, Amazon and Meta continue to ramp up infrastructure investment and Wall Street is becoming noticeably less patient. πŸ—οΈ

What's fascinating is that demand for AI remains incredibly strong. Google's cloud backlog is growing rapidly and customers clearly want more compute. The problem isn't demand it's whether these companies can turn hundreds of billions of dollars of infrastructure into profitable products before shareholders lose their nerve. βš–οΈ


There's another irony too. Google's AI is becoming so good at answering questions directly that publishers including some who've licensed their content to train Google's modelsβ€”are seeing fewer visitors from Search. Better AI may improve Google's products, but it could also shrink the very ecosystem that made Search so valuable in the first place. 🌐


πŸ“š Read More

πŸ” Finimize: Alphabet's solid second-quarter earnings show it's still got the goods
https://finimize.com/content/alphabets-solid-second-quarter-earnings-show-its-still-got-the-goods

πŸ—οΈ Yahoo Finance / CRE Daily: Google's $205B AI data centre bet rattles Wall Street
https://finance.yahoo.com/technology/ai/articles/google-205b-ai-data-center-063942663.html

πŸš— The Guardian: Tesla stock suffers double-digit drop amid AI spending concerns
https://www.theguardian.com/technology/2026/jul/23/tesla-stock-drops-elon-musk

πŸ€– Is Your Favourite Newsletter Actually Written by AI?


TL;DR πŸ‘€ Substack has launched an AI detection tool that lets readers estimate whether newsletters, comments and posts were written by a human or AI. It could expose some creators in the short term, but it may also become one of the internet's first real transparency standards for AI-generated writing.


What happened? πŸ“°

Substack has partnered with AI detection startup Pangram to let readers scan any post, comment or reply over 100 characters and see an estimate of how much was written by a human versus AI. Writers can also add an optional AI Author's Note, explaining how they used AI in creating a piece. Importantly, Substack isn't banning AI-assisted writing or penalising creators who use it, they claim they just wants people to be more transparent. πŸ€–


🧠 So what?

This feels like one of the first genuinely interesting responses to generative AI. Rather than pretending AI doesn't exist, or treating any use of it as cheating Substack is saying: use AI if you want, just don't pretend you didn't. πŸŒ


AI is increasingly replacing everything around the writing research, editing, proofreading, translations and SEO but the hard bit remains having an original idea worth reading. As Substack CEO Chris Best put it, software can do almost everything except "the hard part". That's still the human job. ✍️


We're likely heading towards a world where AI disclosures become as normal as sponsored content labels. Not because people reject AI, but because trust increasingly depends on knowing how something was made. The irony is that newsletters built their reputation on authentic voices escaping algorithmic feeds. Now they may need a badge proving there was actually a human behind the keyboard. πŸ˜…


(And before anyone asksβ€”yes, we use AI. It helps us research faster, spot patterns, challenge our assumptions and catch the odd (or plentiful when it was all manual)ntypo. The Editorial stance, opinions, analysis and questionable jokes are still ours. πŸ˜‰


πŸ“š Read More

πŸ€– TechCrunch: Substack's new tool tells you who's been writing their newsletters with AI
https://techcrunch.com/2026/07/22/substacks-new-tool-tells-you-whos-been-writing-their-newsletters-with-ai

✍️ Pangram: Learn more about the AI detection technology powering Substack's new feature
https://pangram.com/

πŸ‘©πŸΏβ€πŸ’»For the creatorsπŸ‘©πŸΏβ€πŸ’»

πŸ“ˆ The tools behind the techπŸ“‰

πŸ“¦ProductπŸ“¦

πŸ“DesignπŸ“ 

πŸ‘©πŸΏβ€πŸ’»CodeπŸ‘©πŸΏβ€πŸ’»

🏒The business behind the tech🏒

πŸ›οΈTech deal of the weekπŸ›οΈ

A Lenovo slim laptop for under Β£400, with enough ram and a processor to open at least ten tabs on chrome, Take a look here


Link here and check out our other deals too


And view our shop with our whole collection here


πŸ˜…Meme/AI video of the week πŸ˜… (the internet can be savage lol)

🌐Partner Events & Opportunties 🌐

Below are the top opportunities we want to highlight to you this week! If you want to see more, then check out our new website where we have a whole page dedicated to events and opportunities from us and our partners:


https://www.colorintech.org/events

Did somebody say... LGBTQ+ inclusion? 


Well, we did! We are collaborating with Just Eat Takeaway to celebrate the LGBTQ+ community through a panel discussion and workshops designed to help you challenge limiting thoughts so you can become your most fearless self.
We have an incredible panel lined up including: 


 MNEK: Music artist with over 3.8 billion streams and written for superstars like BeyoncΓ©, Zara Larsson and Flo.


 Belton Flournoy: Managing Director of Protiviti and named Top 10 Inspirational Business Leaders at the British LGBT Awards 2020


 Malika: Female popstar who has written for superstars like Leigh-Anne Pinnock and Ayra Starr


The event will also feature workshops, giveaways and time to network with people from Just Eat and the Colorintech community.


Date: Thursday 6th August
Time: 15:30 - 18:30 (with time for networking) UK Time
Where: Central London


 Register to secure your free ticket here (spaces are limited and subject to availability) 


Whether you identify as LGBTQ+ or are an ally who believes in the power of inclusive spaces, you don't want to miss this event! 🌈




πŸ™ŒπŸΎThe latest from the Colorintech teamπŸ™ŒπŸΎ

πŸ˜ƒWhat we are consumingπŸ˜ƒ


🎧Snapchat and music just got closer

πŸ‡«πŸ‡· France bans social media for kids too

🏰 Map of castles and forts

🚝 a 3D shinjuku station


Email Marketing by ActiveCampaign