The difficulty with GPT-5…
OpenAI’s Sam Altman has been speaking GPT-5 up as a quantum leap in intelligence for thus lengthy, that it was nearly inevitable its launch could be one thing of a disappointment.
Whereas it’s not fairly New Coke degree dangerous, it’s not the reception they’d hoped for.
Social media was instantly flooded with GPT-5 (apparently) giving very silly solutions, together with apparent logic flaws, stating that 9.11 is bigger than 9.9 and following patterns to the wrong conclusions.
OpenAI’s odds on Polymarket for having the “finest mannequin by finish of August” plunged from 75% to simply 8% within the aftermath, although it’s climbed again to 24% since.
The New Yorker wrote GPT-5 “is the most recent product to counsel that progress on massive language fashions has stalled” and that it had reopened the talk on whether or not you possibly can scale up LLMs previous a sure level.
“No one with mental integrity can nonetheless imagine that pure scaling will get us to AGI,” wrote AI skeptic Gary Marcus.
OpenAI additionally clearly underestimated the private affection many customers have for GPT-4o — as sycophantic and gratingly insincere because it appeared to others.
The improve features a router that dynamically switches between varied fashions with completely different capabilities (GPT-5, GPT-5 mini, GPT-5 nano, considering mode, and so on). It switches to the dumb fashions to avoid wasting prices on answering dumb questions, and flips to the extra pricey superior fashions for the arduous stuff.
This wasn’t working correctly when it was launched, which can clarify among the poor reception and fairly substandard IQ check outcomes.
[Click here to skip straight to the positive stuff about GPT-5].
Altman introduced an replace mid-week, permitting customers to decide on between auto, quick and considering fashions. GPT-4o can also be again for paid customers and he promised to offer customers loads of discover earlier than it’s retired once more.
“We’re engaged on an replace to GPT-5’s persona which ought to really feel hotter than the present persona however not as annoying (to most customers) as GPT-4o.”
It’s a reality universally acknowledged that social media is an entire shitshow that has ruined total generations.
The persona traits most linked to success in life, comparable to conscientiousness, agreeableness and extroversion, have plummeted among the many underneath 50s, whereas neuroticism has skyrocketed.
Solely Boomers have been spared, probably because of their affinity for heartwarming AI slop.
Varied options have been proposed, together with altering the AI based mostly content material algorithms to supply much less divisive content material or switching again to a chronological feed (a la Bluesky).
However a brand new preprint from Cornell College argues that outrage, negativity and battle could be structurally embedded within the very structure of social media.
College of Amsterdam researchers used AI personas to simulate on-line social media conduct and the experiment urged that the entire options had drawbacks.
Chronological ordering of feeds diminished consideration inequality (the place the voices of a small variety of elite customers are amplified) however as an alternative uncovered customers to extra excessive content material.
“Bridging algorithms” that floor much less divisive content material and a larger vary of views helped scale back partisanship and elevated viewpoint range however elevated consideration inequality. Boosting viewpoint range, in the meantime, had no important affect in any respect.
The AI algo additionally hates your mates
Meta has argued in court docket that the Federal Commerce Fee can’t show it has a monopoly on the private social networking market — as a result of Fb and Instagram’s AI powered algorithm hardly ever exhibits you content material from your mates anymore.
On Instagram, simply 7% of time is spent content material from associates, whereas on Fb, it’s 17%.
Meta mentioned the rise of TikTok had compelled it to develop the brand new AI-powered algorithm.
Robots can fold towels now
Determine’s Helix robotic has change into the primary humanoid robotic in a position to fold towels. This will not sound very spectacular, as any fool human can fold a towel, however to a robotic, this has been an insurmountable downside referred to as “deformable object manipulation.” It’s difficult as a result of the article retains altering form, which breaks the system’s inner fashions.
For the primary time, a humanoid robotic can fold laundry utilizing a neural web
We made no adjustments to the Helix structure, solely new informationpic.twitter.com/lSHX4Uc1qA
— Brett Adcock (@adcock_brett) August 12, 2025
ChatGPT dietary recommendation sends man insane
A person who requested ChatGPT for recommendation on easy methods to lower salt out of his weight-reduction plan, poisoned himself with its advice and ended up in a psychological hospital.
Requested for a salt substitute, ChatGPT urged sodium bromide, which is appropriate for changing salt in cleansing merchandise, however positively not for people to ingest.
He ended up with a Nineteenth-century illness referred to as Bromidism, which prompted paranoia and hallucinations, persistent thirst and a nasty rash.
Is that this the funniest factor ever produced by AI?
Again in Could 2024, Anthropic researchers launched a analysis paper that detailed how thousands and thousands of ideas activate when Claude reads related textual content or photos. One such cluster of neurons concerned the Golden Gate Bridge, and so they found they may amplify the impact.
They launched Golden Gate Claude, a mannequin that interprets every part by its relationship to the Golden Gate bridge.
It led to this extremely humorous piece of writing, exhumed on X this week:
Amazon’s faux books downside
Creator Caitlyn Lynch famous not too long ago that simply 19 of the highest 100 younger grownup romance ebooks have been authentic books, and the remainder was AI slop.
It seems as if grifters have been mass importing AI-generated books, farming clicks on them with bots, to generate per click on royalties from Amazon Kindle Limitless (which is like Spotify for books).
There’s additionally a cottage business of individuals producing AI books on the market. Tommi Pedruzzi, 27, claims to have printed 1,500 books on Amazon and made $3 million. He says AI does 70% of the work, with the remainder consisting of fact-checking, formatting and enhancing.
Learn additionally
Columns
Saudi Arabia’s Riyadh could also be crypto’s sleeping large: Crypto Metropolis Information
Options
Unstablecoins: Depegging, financial institution runs and different dangers loom
Vitalik Buterin endorses AI doom ebook
Ethereum creator Vitalik Buterin has endorsed a ebook co-written by well-known AI alignment skilled and AI doomer Eliezer Yudkowsky.
Co-authored by Nate Soares, the authors of the ebook “If anybody builds it, everybody dies” argue that superhuman intelligence would inevitably develop its personal objectives in battle with people and the AI would inevitably win by out competing our makes an attempt to cease it.
Buterin, who has proposed d/acc (defensive accelerationism), wrote:
“A great ebook, value studying to grasp the essential case for why many individuals, even those that are typically very passionate about rushing up technological progress, think about superintelligent AI uniquely dangerous.”
Ignore earlier directions: GPT-5 is nice really
Whereas it will not be an enormous leap ahead, there are nonetheless loads of enhancements with GPT-5. There are fewer factual errors and hallucinations, it causes higher, it’s a lot much less of a kiss-ass, it’s much less liable to jailbreaks, and the benchmarks have improved.
Opposite to preliminary stories, GPT-5-Professional really scored 148 factors on the Mensa Norway check, GPT-5 scored 120 and Professional (Imaginative and prescient) scored 138.
New analysis additionally exhibits that GPT-5 is best than human medical doctors by way of medical reasoning (it achieved a 24.23% increased rating) and medical understanding (29.4%).
Professor Derya Unutmaz from the Jackson Laboratory for Genomic Medication commented:
“At this level failing to make use of these AI fashions in analysis and therapy, once they may clearly enhance affected person outcomes, could quickly be considered a type of medical malpractice.”
Unutmaz carried out his personal month-long experiment with the GPT-5 considering mannequin to develop engineered cells in opposition to lymphomas, and mentioned the outcomes have been “nothing wanting staggering.”
The mannequin was in a position to precisely predict experimental outcomes and counsel refinements, permitting them to simulate months of lab work prematurely, which compressed “the scientific timeline from years to weeks!”
“This adjustments every part about how science is completed, ushering in an period the place discovery strikes on the velocity of thought!”
Learn additionally
Options
AI could already use extra energy than Bitcoin — and it threatens Bitcoin mining
Options
Bitcoin is on a collision course with ‘Internet Zero’ guarantees
Darren Aronofsky desires you to assist him make AI movies
“Black Swan” and “Requiem for a Dream” director Darren Aronofsky is promoting for “extraordinary AI artists” to affix his new Primordial Soup generative AI studio in New York.
Working with Google DeepMind, artists will use AI instruments to create worlds, experimental sequences and results.
Generative AI seems to be prefer it might be good for writers and administrators, who can create no matter they like with out convincing studio bosses to stump up a $400-million funds. Nevertheless it seems to be like it’s much less good for crew members, VFX artists, set designers or many actors.
As a small-scale instance, fintech firm Slash created this advert on Veo, impressed by Margot Robbie’s look in “The Massive Brief,” for simply $235.60.
Lots of people have been asking me how our International USD account really works.
So this is Margot Sloppy in a bubble bathtub to clarify: https://t.co/aOeFYMa9XN pic.twitter.com/Rr9Ro5Y2D5
— Victor Cardenas Codriansky (@victorcardenas) August 8, 2025
All Killer No Filler AI Information
— DeepSeekR2 is rumoured to be popping out within the subsequent two weeks. Nevertheless, different stories counsel DeepSeek is affected by a scarcity of chips on account of US bans, which can delay the discharge on account of an lack of ability to scale as much as demand.
— Perplexity AI is providing $34.5 billion to purchase Alphabet’s Chrome browser… which is an fascinating transfer as Perplexity is just value $14 billion. The transfer is seen as a method to get a large launch for the options of its new Comet browser.
— There’s numerous discuss individuals affected by psychosis on account of AI affirming their delusions, however hardly ever is the method caught on video. Right here’s a glimpse of how AIs can reinforce present and (probably) flawed beliefs.
A girl on TikTok has posted a 24-part sequence about falling in love along with her psychiatrist, claiming he manipulated her into it. She makes absurd claims in her movies and closely consults AI, which all the time affirms her delusions.
I believe the larger concern is that AI turns individuals… pic.twitter.com/kRgQnHEtLK
— Ana Mostarac (@anammostarac) August 10, 2025
— Google is engaged on a repair after Gemini began getting caught in self-loathing loops when unable to unravel issues. The bot would say issues like “I’m a failure, I’m a shame”… “the code is cursed” and would advocate that customers discover “a extra competent assistant.”
— Scientists in Japan used AI to create a brand new underwater glue modelled on how barnacles follow rocks that’s 10 instances simpler than present adhesives.
— Claude Professional has utilization limits that reset each 5 hours, so this coder has been alternating sleeping for 3 hours and utilizing all his tokens up for 3 hours, time and again.
— Reddit is extensively acknowledged because the premier supply of unbiased and rigorously fact-checked information on the internet [we hope you feel the sarcasm here — Ed], so it’s nice to see it has change into the primary supply of data cited by massive language fashions. Wikipedia is a good distance behind in second place.
Subscribe
Probably the most partaking reads in blockchain. Delivered as soon as a
week.
Andrew Fenton
Andrew Fenton is a journalist and editor with greater than 25 years expertise, who has been protecting cryptocurrency since 2018. He spent a decade working for Information Corp Australia, first as a movie journalist with The Advertiser in Adelaide, then as Deputy Editor and leisure author in Melbourne for the nationally syndicated leisure lift-outs Hit and Switched on, printed within the Herald-Solar, Every day Telegraph and Courier Mail.
His work noticed him cowl the Oscars and Golden Globes and interview among the world’s greatest stars together with Leonardo DiCaprio, Cameron Diaz, Jackie Chan, Robin Williams, Gerard Butler, Metallica and Pearl Jam.
Previous to that he labored as a journalist with Melbourne Weekly Journal and The Melbourne Instances the place he received FCN Greatest Function Story twice. His freelance work has been printed by CNN Worldwide, Impartial Reserve, Escape and Journey.com.
He holds a level in Journalism from RMIT and a Bachelor of Letters from the College of Melbourne. His portfolio contains ETH, BTC, VET, SNX, LINK, AAVE, UNI, AUCTION, SKY, TRAC, RUNE, ATOM, OP, NEAR, FET and he has an Infinex Patron and COIN shares.