News
- GPT-4 predicted personality scores months before any human ever answered a question
- A Bosch oven manual was enough for GPT-4 to invent personality traits
- DSM-5 predictions hit 0.71 accuracy; astrology predictions somehow hit 0.85 accuracy
Human personality has traditionally been measured by asking people questions and comparing their answers across established psychological frameworks.
A new study published on iScience suggests GPT-4 can go further by estimating how people collectively respond to personality questions before answering begins.
Researchers Rotem Monsa, Aviv Zohar, and Shahar Arzy of the Hebrew University and Hadassah Medical School tested whether the model could infer patterns in human responses from language alone, rather than merely generating plausible questionnaire wording.
GPT-4 predicted responses before humans answeredThe team used GPT-4 to create personality questionnaires from two very different textual sources.
One questionnaire drew on the DSM-5, the standard clinical manual used worldwide by psychiatrists to diagnose personality disorders with precision.
The other version pulled statements from an astrology textbook that assigns particular personality traits to each of the twelve zodiac signs.
"We wanted to choose texts that describe human personality in very rich detail but also sit on opposite ends of a spectrum in terms of scientific grounding," said Rotem Monsa, lead author of the study.
“The DSM-5 was refined through decades of clinical research and is the standard diagnostic manual in clinical psychiatry, known worldwide.”
Both questionnaires were then given to 600 participants alongside the Big Five Inventory, the most validated personality measurement tool available.
Before collecting those responses, GPT-4 estimated average scores for individual questions using a five-point Likert response scale.
This gave the researchers a way to compare predictions against subsequently collected evidence.
The model's estimates correlated strongly with actual participant averages, reaching 0.71 for DSM-derived questions and 0.85 for astrology-derived questions.
This decent internal consistency for DSM-5-based questionnaires suggests that traits that typically appear together in real life also correlated in responses.
The astrology questionnaire, by comparison, showed a noticeably weaker internal consistency across most of the personality traits it measured overall.
GPT-4 also estimated relationships between individual questions, producing correlations of 0.74 and 0.69 for DSM and astrology questionnaires, respectively.
The researchers therefore argue that GPT-4 appears capable of capturing population-level response tendencies embedded within language and questionnaire content.
However, they specifically caution that predicting average responses does not establish that the resulting questionnaire accurately measures individual personality differences.
(Image credit: iScience)Testing the limits with an oven manualThe researchers also tested whether GPT-4 would generate personality-style questions when given material containing no explicit personality information.
They supplied the model with a Bosch oven manual and a literary description of a landscape from The Lord of the Rings.
GPT-4 still produced questionnaire items resembling personality assessments, even though neither source was designed around psychological traits.
That result suggests the model has learned strong associations between language patterns, personality descriptions, and familiar questionnaire structures.
Yet the researchers found that unrelated source material could produce items that were semantically weak, narrow, or poorly connected.
This distinction matters because a convincing question does not necessarily measure the psychological characteristic that researchers intended to examine.
The study also found important differences between the questionnaires when researchers examined their underlying statistical structures.
The Big Five questionnaire most closely matched its expected organization, while the DSM and astrology questionnaires showed weaker confirmatory factor-model fits.
Astrology-derived items also failed to preserve their intended elemental structure, instead forming broader patterns resembling established personality dimensions.
The results nonetheless showed comparable predictive performance across many life outcomes, although astrology-based measures performed worse for mental-health outcomes.
For now, the evidence supports a narrower claim: GPT-4 can anticipate aggregate human response patterns from textual information with measurable accuracy.
“Given that personality traits are reflected in language, LLMs may have learned the structure of human personality as a natural byproduct of their training,” Monsa added.
“So, while they were not taught specifically psychology or personality theories, these are already embedded in the language that LLMs learn from.”
Mild spoilers for Stuart Fails to Save the Universe episode 4 ahead.
Four episodes into Stuart Fails to Save the Universe on HBO Max and we've seen four surprise Big Bang Theory characters return: Raj (Kunal Nayyar), Penny (Kaley Cuoco), Bernadette (Melissa Rauch) and Amy (Mayim Bialik).
With one new face per episode, you might be wondering whether the remaining six episodes in the series will continue the trend.
Most notably, Sheldon (Jim Parsons) and Leonard (Jonny Galecki) have yet to make an appearance — and considering they built the machine that Stuart (Kevin Sussman) broke that got him into this mess, it feels fitting that they should reappear. Howard (Simon Helberg) has also yet to reappear.
However, a return isn't necessarily guaranteed, especially if the creators want Stuart Fails to Save the Universe to be picked up for a second season (which they definitely do). Why give us what we want now if they can drag it out?
But after speaking to Chuck Lorre, Bill Prady and Zak Penn, it turns out the reason for the season 1 cameos is actually boringly straightforward: it all came down to who was available to film.
'It was what was necessary for the storytelling'(Image credit: HBO Max)"It was availability and what was necessary for the storytelling" Lorre responds when I ask him about this season's cameos. "It was wonderful to have Melissa play a demonic character, for example. That just made sense."
Penn adds "Who we had something good to write for," with Prady agreeing "I also loved finding out that Amy is married to Golden Age Flash. That made me very happy."
Perhaps we should be focusing on these Easter egg details, rather than what cameos may or may not appear. Why was Golden Age Flash such a good choice? Because he's Sheldon's favorite DC superhero, and the reason why, nearly every time we see him, he's sporting the iconic T-shirt with the Flash logo.
I don't want to brag, but I called this the moment that a Stuart Fails to Save the Universe DC crossover was revealed. But as for continued cameos, I'm not as confident.
It's probably more likely that we'd see every variant of Flash known to man before Sheldon shows up, because TBBT feels like something that's resolutely a part of Parsons' past.
However, adult Sheldon did show up for the Young Sheldon season 7 finale, so anything is possible. But if we're going there, I need George Cooper to be resurrected more than I need to see adult Sheldon again.
In fact, now that we've had Christine Baranski appear as herself, let's get Laurie Metcalf (who played Mary Cooper) back on the scene instead. HBO Max... call me.
AeroPress coffee makers are going cheap right now, in a whole rainbow of colours, and you'll find the complete set right here. Almost every colour is 20% off at Rave Coffee right now, except for the rare green model, which is out of stock at most retailers but available with a slightly more modest 16% price cut at Coffee Omega.
These colourful coffee makers are super lightweight, and even more versatile than you might have realised. Did you know that you can use them to make cold brew coffee in just two minutes, for example? No need to spend 24 hours waiting for your drink to brew, and then make a mess straining it through a cheesecloth. Just follow the instructions from AeroPress's official site — and don't forget to stir.
Want to get even more creative? How about using your AeroPress to make an Irish coffee, a carrot cake latte, or a matcha coffee fusion? Once you've picked up your new coffee maker, the sky's the limit.
Colorful AeroPress deals AEROPRESS Coloured AeroPress Coffee Maker — Blue AEROPRESS Coloured AeroPress Coffee Maker — Red AEROPRESS Clear AeroPress Coffee Maker AEROPRESS Coloured AeroPress Coffee Maker — Pink AEROPRESS Coloured AeroPress Coffee Maker — Purple AEROPRESS Coloured AeroPress Coffee Maker — GreenThe Google Pixel 11 range is here with preorders officially underway in Australia, and we’ve covered all the best offers from both retailers and telcos across the country.
While there are plenty of great options out there, for our money, the best preorder deal comes from Vodafone. The telco is quite literally giving away a Sony PlayStation 5 Slim console and an extra DualSense controller for free when you preorder the Google Pixel 11 Pro XL until August 19.
Given that the PS5 Slim has seen multiple price increases since launching in 2020 and discounts are getting harder to come by, this deal is hard to pass up for any gamer looking to upgrade their phone and console at the same time, all from as low as AU$103.30p/m for a 36-month contract.
Google Pixel 11 Pro XL (256GB) | Vodafone Small Plan | 65GB | 24 months | AU$132.45p/m
This offer is available whether you sign up for a 12-, 24- or 36-month plan with the Google Pixel 11 Pro XL. Vodafone is offering the 256GB in only three colours — Canyon, Olive and Obsidian — and the 512GB model is only available in Obsidian. The Small Plan includes 65GB of mobile data, unlimited national calls and texts, unlimited international texts and 5GB of daily roaming for an extra AU$5 per day.View Deal
The Google Pixel 11 Pro XL brings some notable upgrades from its predecessor, headlined by the new Tensor G6 chipset, a brighter display with 3,600 nits peak brightness (up from 3,300 nits) and a new LED notification light in the camera flash module called HiLight.
The cameras also get upgrades, with the main shooter having a new sensor designed to improve low-light performance, as well as boosting the zoom capabilities up to 120x, up from 100x in the Pixel 10 Pro XL.
It’s not all good news, unfortunately, as the base 256GB model only gets 12GB of RAM instead of 16GB in last year’s model (despite the price increase from AU$1,999 to AU$2,099), but it was to be expected given the current climate of component shortages.
Fortunately, this bundle with a PS5 thrown in for free could go some way to softening the blow, and is a deal that you can’t miss if you’re in the market for a new Pixel phone and a new gaming console. The convenience of a monthly payment instead of shelling out AU$3,108 for both devices and a spare DualSense controller makes it an even more compelling buy.
Be quick, as Vodafone is only running this promotion until Wednesday, August 19.
I didn't set out to build a Breakout clone, but that's what I — or really Claude, under my direction — did, and it took less time than it takes to watch a half dozen cat videos.
Here's how it started. Claude was having issues. Anthropic's powerful and popular generative AI platform was apparently going up and down, and some were complaining, but when I checked, Claude told me, "I'm doing well, thanks for asking!" I realized, though, that if the problems were intermittent, I might need a tougher task to test Claude's overall stability.
More than a year ago, I worked with Claude, or rather guided Claude through the process of recreating the early 1980s arcade game classic: Asteroids. The final game, which Claude exported as a shareable artifact, was the product of hours of guidance and back-and-forth between me and the AI. It didn't start off great, but with each prompt, we iterated, and the game improved until I had a rather robust and true-to-the-original Asteroids knockoff.
With that in mind, I tasked Claude AI, running the free Sonnet 5 model at "Medium" level, with creating a Breakout clone.
Breaking outBreakout is one of my favorite games, something I've played on early consoles, PCs, and even my old BlackBerry, where "Brick Breaker" was one of the few preloaded games.
The game rules are pretty simple. There's a multi-level brick wall (usually eight bricks deep) at the top of the screen, a paddle at the base, and a bouncing ball you bounce off the paddle to break down the brick wall.
For my prompt, I tried to describe, in some detail, my game, and was not coy about it being an online version of Breakout:
Let's make a quick game. I want an online version of the old Breakout game, but let's replace the bricks with robots that are always slowly marching down toward you (if they reach you, it's game over). The ball that you bounce off your side-sliding controller is a gear that breaks up the robots easily in the first level. In the second level, the robots are shinier and harder, and the gear takes three or more hits to break and make a robot disappear. If you clear the level, the gear turns into a titanium ball that bounces faster but is more powerful. Controls are either keyboard/arrow based and mouseable. Want this to result in a web-based, shareable game, so a link/artifact I can share online.
(Image credit: Claude)That's all the direction I gave it. I didn't describe colors, the initial speed of the ball, or even how quickly the robots would march down the screen. I didn't get into details about the look of the robots, instead relying on generalities and some material descriptions to help guide Claude.
It took me a minute or two to write the prompt and around three minutes for Claude to deliver the results, which it cleverly called "Gearbreaker".
I got a brief description of gameplay and levels (there are just two for now) :
"How it plays:
Level 1 — matte robots, one hit each from the spinning gear ball. They march downward on a timer; let them cross the dashed red hazard line and it's instant game over.
Level 2 — shinier robots (metallic gradient, cyan eyes) that soak up 3 hits, shown as pips along the bottom of each unit. Same gear ball, tougher targets.
Clear level 2 → the ball forges into a titanium core: bigger, faster, hits for 2 damage instead of 1. Levels keep scaling procedurally after that (more rows, faster march, tougher units), so it runs endless with a local high-score save.
Controls: arrow keys or mouse to slide the paddle, Space or click to launch/pause/restart. Basic touch support too."
On the right side of the screen, there was my playable game. There's also the artifact publish button, which I chose to use so I could share the game with you.
It's a stunningly complete render of a completely playable and, in some ways, challenging Breakout clone, and it took five minutes. I'd call the design clean and efficient. The bricks have a robot look, and the ball is a blue spinning gear (in the first level). The movement and physics are basically perfect, and there's enough on-screen instruction to easily grasp gameplay.
Just the beginning(Image credit: Claude)It's just another mind-boggling example of the incredible pace of AI development. While I had expected a playable game, I assumed it would be missing enough crucial components to make it dull or uninteresting — something similar to my earlier AI game development projects. Instead, I got something that displays a clear understanding of the purpose of such a game: challenge, entertainment, and replayability.
Using my mouse, I played through a couple of times, failing to break all the bricks in Level One and ending the game. I realized I had to take Claude's output more seriously, and, with some effort, as I watched the robot bricks steadily march down my screen, I finally cleared all the bricks in Level One and then started in on the tougher Level Two. As I had instructed, the robots on the second level looked and reacted differently. It now took at least three direct hits to make them disappear. I failed and already want to play again.
(Image credit: Claude)This is by no means the wildest example of the AI development process or vibe coding; it's just me prompting Claude to code a complete game with the barest of instructions.
Claude and these models cannot do this on their own, but it does make you think about the agentic AI that's now being unleashed on the world. It's more powerful and autonomous. I could imagine someone asking one to start work on recreating all the popular video games of the last half century. The only limit would be time and compute tokens.
Speaking of which, Claude just informed me that this Breakout task used up about 90% of my credits for the day. I think it was worth it.
Take the game for a spin and let me know what you think in the comments below.
In today’s supply chain, a week is not merely a long time. It can be the difference between protecting margins and absorbing avoidable cost; maintaining availability and losing sales; or preserving customer trust and explaining why another commitment has been missed.
Geopolitical tensions, tariff changes, economic uncertainty and supplier instability are no longer occasional interruptions to an otherwise predictable operating environment.
They increasingly overlap, interact and move faster than traditional planning cycles.
The external risk picture supports that conclusion. The World Economic Forum’s Global Risks Report 2026 identifies geoeconomic confrontation as the leading risk for both 2026 and the period to 2028. Half of the experts surveyed expect the global outlook over the next two years to be turbulent or stormy.
This is creating a decision-speed challenge at the heart of global commerce.
Research with supply chain leaders, found that only 20% can develop and deploy a response to a geopolitical disruption within 24 hours. A further 38% require more than a week.
During those seven days, transport costs can change, capacity can disappear, inventory can become stranded and competitors can secure alternative sources of supply. By the time a response has passed through every functional review and approval, the original assumptions may already be obsolete.
The critical question is not whether an organization can see disruption. It is whether it can convert that signal into an executable enterprise decision while there is still time to influence the outcome.
Visibility is not the same as readinessFor many years, supply chain transformation focused on improving forecasts, optimizing individual functions and reducing cost. Those disciplines remain important, but many of the operating models surrounding them were designed for a more predictable world.
Today’s disruptions cut across procurement, manufacturing, logistics, inventory, commercial planning and finance simultaneously. Yet many organizations still manage them through separate systems, functional metrics and sequential decisions.
Consider a tariff change. It may require the business to reassess sourcing, production, inventory deployment, pricing, customer allocation and margin exposure. If each function develops its own answer before the enterprise reconciles the trade-offs, valuable time is lost.
A dashboard may reveal the disruption quickly, but it does not determine which customers to prioritize, what financial exposure is acceptable or who has the authority to act.
Visibility without decision rights is simply faster awareness of the same problem.
Confidence is becoming polarizedThe research also reveals a decline in overall confidence. Only 66% of leaders now consider their supply chains ready for the future, compared with 73% in 2025.
46% describe themselves as highly optimistic about their supply chains. These leaders are also more likely to report shared cross-functional KPIs, integrated data and stronger end-to-end visibility.
Optimism itself does not create better performance. Rather, confidence appears to reflect the capabilities these organizations have already built.
Less optimistic organizations are 2.3 times more likely to experience slow data sharing and integration, 2.5 times more likely to operate in functional silos and four times more likely to describe their supply chains as disjointed.
The divide is therefore not simply technological. It is organizational and operational. Connected businesses are better positioned to develop a common understanding of events, evaluate enterprise-wide consequences and align teams around one response.
From unified data to unified decisionsUnified data platforms are now the most widely adopted new technology in the survey, deployed by 51% of organizations. Adoption rises to 64% among the more optimistic cohort, compared with 40% among less optimistic leaders.
That foundation is essential, but it is only the beginning.
The real value emerges when common data supports connected decisions across planning and execution. This enables organizations not only to identify disruption, but also to assess its end-to-end impact, evaluate alternative responses and orchestrate action across the network.
Technology provides the foundation, but achieving decision speed also requires shared enterprise outcomes, clearly defined decision ownership, agreed intervention thresholds, pre-modelled scenarios and the authority to act.
European businesses are already adapting. European Commission survey evidence found that 27% of EU firms had changed, or planned to change, their strategies in response to tensions, disruption or policy changes in foreign markets. Among those adapting, 38% were changing sourcing or destination countries, while 22% were increasing inventory buffers.
Resilience is no longer an abstract ambition. Businesses are actively reconfiguring supply networks—but every adjustment carries consequences for cost, working capital, service and risk. The faster those trade-offs can be understood across the enterprise, the more choices leaders retain.
In grocery and consumer goods, the response window is narrower still because shelf life, availability and promotional commitments leave little room for delay.
AI can accelerate decisions - but foundations determine valueAI will play an increasingly important role in reducing the time between signal, insight and action. Machine learning and predictive AI are already used by 45% of respondents, with another 29% implementing them. Generative AI adoption has doubled from 12% to 24%, while agentic AI remains at an earlier stage, with 8% reporting live deployment.
AI can continuously monitor conditions, identify exceptions, evaluate scenarios and recommend responses at a scale human teams cannot match. When connected to planning and execution, it can also help translate those decisions into coordinated action across the network.
However, autonomy is not a shortcut around unresolved transformation challenges. AI operating across fragmented data, conflicting objectives and unclear decision rights may accelerate activity without improving outcomes.
Before giving AI agents greater scope to act, leaders must be confident in the data, guardrails, accountability and operating model surrounding those decisions. The question is not simply, “Can the agent act?” It is: “Should it act, within what boundaries and in pursuit of which enterprise outcome?”
The new measure of resilienceNo organization can predict every disruption. A more practical measure of resilience is how quickly the organization can understand what has changed, determine what matters, make the necessary trade-offs and mobilize a coordinated response.
That is the real 24-hour test.
The organizations pulling ahead are not simply those with more information. They are those that can connect intelligence, decisions and execution quickly enough to act while choices still exist.
In an increasingly volatile world, resilience will not be determined by who sees the disruption first. It will be determined by who can make - and execute -the best decision before the window to respond closes.
We list the best data visualization tools.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
- Insomniac Games explains why Logan's suit in Marvel's Wolverine heals itself
- Head of technology Mike Fitzgerald says there are "some practicalities to it" and "you can’t play through most of the game wearing just shredded rags on his body"
- He adds that the team invested a lot of time in showing damage progression to suits
Insomniac Games has explained why Logan's suit in Marvel's Wolverine regenerates after fights, and its reason makes a lot of sense.
If you've watched any of Marvel's Wolverine trailers, you've probably noticed that after bloody battles, Logan's fancy hero suit heals itself of any combat damage.
Wolverine can heal his injuries very quickly thanks to his innate mutation, and since his skeleton is coated in an alloy called adamantium, he's near-indestructible. Initially, fans thought this might be why his suit was regenerating in the game, or perhaps it was even nanotech; however, Insomniac has settled the matter, explaining that it's simply about keeping the superhero vibe intact.
"I think mostly there’s some practicalities to it,” head of technology Mike Fitzgerald said in an interview with IGN. "You can’t play through most of the game wearing just shredded rags on his body. So we find the right balance between clothing that reflects the gameplay you’re doing and this brutal combat and the realities of that, and then finding the right moments to have it restore itself so you get that right cinematic feel of Wolverine."
Senior project director Jess Reiner-Reed also commented on finding the right balance, adding, "Early on you would be playing through missions and your clothes are shredded for way too long. You don't feel as powerful as you're playing through. So yeah, we definitely found that right balance was the right thing to say."
While Logan's suit will heal after fights, Fitzgerald explained that the team still had to show damage progression across his different suits and enemies, as well.
"We invested a lot in how damage is reflected on him in his suit, on enemies and their suits," Fitzgerald explained. "And then as you attack them, the blood sprays coming off them are some really interesting fluid tech. So it's not just pre-baked visual effects. These are actual blood particles coming out.
"They hit the ground and make the exact splatter mark they would make if they land on Logan. They leave blood there. If he walks through water, it washes off and leaves blood swirls in the water."
Marvel's Wolverine launches on PS5 on September 16. Insomniac recently confirmed that it will be "fully playable" on disc following Sony's decision to end physical disc production.