Post
My critique is a bit more pointed. The 3.5 flash model isn't a reasoning model and it's the lowest common denominator so it will make much worse mistakes. Higher end models are less prone to this. Bleeding edge models are solving major mathematical proofs.
techcrunch.com/2026/05/20/o...
So the AI that most people will interact with on a daily basis is inferior and prone to far more mistakes. The limitations are magnified because most people will not pay for access to advanced models. And once the bubble pops and costs soar, the disparity will broaden.
i had this absurd google search experience a couple days ago
Hilarious.
Is Google committing brand suicide?
I hope so.
And the tech bros think they can replace us anytime soon with this shit
The fucked thing is they will, and when it crashes and burns, they'll hire us all back as " AI quality control". The same thing happened to a lot of translators when translation software started to take off. They lost their jobs, the software sucked, they got hired back cheaper as "copy editors"
I assumed this was a joke but I tried it anyway.
I was a history major, so take what I say with a grain of salt, but that result may not be correct. Huh.
Idk I think we should destroy the planet for this technology! Think of the potential!
Lol, imagine they actually archieve their goal of AGI and instead of superintelligence it has a personality, is socially awkward and bad at everything.
"Yes, that is incorrect." Lmao
Sometimes it knows math, sometimes it doesn't. Sometimes it knows English, sometimes it doesn't. You just never know.
You’re being VERY picky. It is well known and accepted that AI has to add a bit to cover the energy and water that it is using.
Play fair!!!!
Tried the same thing and it said it was correct, but then in the same response checked again and said it was wrong. Not sure why it did this and still included the incorrect part at the top when it all happened in less than a second.
Axl M
@captain-bleuten.bsky.social
· 2d
"Imagine, you take the #1 search engine in the world known for being fast and accurate and you turn it into a chatbot which gives you less accurate answers while using 10 times more computing power"
Ah so the problem here is that you didn’t prompt it to “make no mistakes”. Try asking the AI to not be incorrect. This is a foolproof method that is sure to work
It might be working fine, just working on the average American math
Ecosia, duck duck go, mojeek, qwant, no reason why I'm writing these words here
Ditch Google
This is what happens when you apply stochastic algorithm to a discreet problem. Even if "better" AIs don't make the same mistake, they have the same underlying issue. We managed to program computers to fail at one task they were singularly designed to be better than any human being. Truly pathetic.
The confidence in that exclamation point is doing some heavy lifting here 😅 There's something deeply unsettling about algorithms deciding what counts as "truth" — like we're outsourcing epistemology to black boxes.
It's always funny when you correct it, "That's a good catch....", "You are correct....", etc... Then it either corrects itself briefly or gets something else completely wrong.
August (Fuck Putin)
@augustclark.bsky.social
· 16h
With the recent Google update in full swing, it’s worth knowing that you can use -ai after your searches in order to get rid of the AI summary.
Also be sure to use a VPN or data protection browser like Brave in order to get rid of trackers and avoid ads.
I was doing my taxes, I gave NotebookLM 48 invoices and asked it to calculate the total.
These were PDF's with all the numbers in the metadata.
No OCR needed.
I doubled checked the numbers in a spreadsheet.
If I had used the number the word vomit generator gave me, I would have committed tax fraud.
FAKE NEWS:
Your forgot the $20 tip for the great AI service.
BEEP BOOP BOP. [MUST GIVE AN RESPONSE NO MATTER WHAT]
At least this crap doesn't burn the planet.
Or ruin a once incredible tool to search anything amongst human knowledge.
Wait.....
AI is not for math. It's not for deterministic workloads in general. The problem has never been the fire, but the people who can't help but set themselves on fire with it. Though the AI marketing teams should probably take some splash damage on this one. Let me show you this spoon that forks... :/
After a PRP procedure where the Dr explicitly said don’t eat/take anything that would help with inflammation, I asked GAI if the Blue Bell Protein ice cream had anti-inflammatory properties and it basically said “No, of course not, moron”. After eating 1/2 pint, I saw the blaring ALA label.
I am still happy that I got my own local AI running, it's good for a lot of things but is a bit dated coming from 2023. Even text versions use a LOT of local compute.
Karl Popper said math is a tautology and completely unfalsifiable so if you want to be a logical science man you must simply accept that this math is completely true and completely false at the same time. ;3
I believe this is being trained on trumpettes logics of how math works
Wrong tool for the job. You want to do math, Claude Code or something agentic with python or similar will plow through stuff like this. You want to see the power of google’s ai, try deep research. That shit will write you a research paper in an hour or so. The problem isn’t the tool, it’s the user.
Are we seriously debating about which AI software to use, to make sure we get the correct answer to a calculation an 8-year-old could do, tolerating errors we wouldn't have admitted at the rise of first calculators ?
And to be fair no, we are not debating which to use. The people who have learned this new tool are busy using it. The people who haven’t are job hunting and complaining about AI.
You should read up on what the Luddites were actually objecting to. It wasn’t the stocking frame. The their economic arguments ring true today in the context of LLMs.
The scariest thing about this to me is you are logged in to your Google account while searching.
Also your profile has your picture!?
The worst part is that it used to just pull up a calculator.
Seems to be working fine.
I'd be curious to know what is after the (bec... ) in your query.
It's funny you assumed it was "bec...", as we only see "b...". Also she already posted the full querry, it was breakdown.
Asked if trump is a criminal. So, it’s a broken clock.
The problem I see here: ai has not reached the point where 1 model can be trained on and to do everything. Ai google seach is an AI for seach, not math.
Of course you can use a calculator but im sure you just wanted to show how ineffectual ai is.
If you want an ai model good at math: Wolfram Alpha, ChatGPT, PhotoMath, and Symbolab. Each offer unique features for solving problems and providing explanations.
I am always interested in dunking on AI, but a better example might be too show examples where search is wrong.
I googled a query about buying property in a certain country and in the same AI summary, it contradicted itself
At the same time, the AI summary is getting better
It's inevitable when "AI" is neither artificial nor intelligent, but merely a mashup up plagiarized material, images and videos they're covertly scraping off the internet, good & bad. I know people punking it by boosting fake stories they make up. The punked stuff is starting to show up now.
Every math teacher right now
GIF
This was me earlier actually trying to find something out:
bsky.app/profile/zzyz...
⏱📋David Steinberg⏱📋
@zzyzx.bsky.social
· 18h
Try to figure out who played drums on Crazy on You, Google's AI search gives the right link to wiki but the wrong info. The link specifically says, "Kat Hendrikse – drums"
Mike Derosier played on some tracks but not that one.
How many people are going to actually going to doublecheck.
It's because they used Abbott & Costello to train their AI model
Now let's ask Google AI what 9 + 10 is 🙃
It has to steal the math from someone and this problem just may not be out there yet.