@GerardPinzone
Joined November 2011
New Post: Low Carb Kaboom Ice Cream gpinzone.blogspot.com/2023/1…
3
1,644
AI slop is dead. The future is now.
#StarWars #drama #horror Anakin's Revelation: The return to the Light. full #ai #movie ⭐️⭐️⭐️⭐️⭐️
5
You’re not gonna believe this guys, but that thing that doesn’t happen happened again.
JUST IN: Army veteran jailed for 30 days over a shooting he did not commit after investigators relied on Flock surveillance data to identify him as a suspect.
80
2,873
10
20,680
271,234
Gerard Pinzone retweeted
"LLMs can't do math" "LLMs can't do new math" "LLMs can't do interesting math" "LLMs can't do open-ended interesting math" "LLMs can't do affirmative open-ended interesting math" "LLMs can't do highly abstract, non-easily-verifiable affirmative open-ended interesting math"
So the point of the stochastic parrot argument is that LLMs have zero understanding of language or anything else, they are just regurgitating training data. This hypothesis has been thoroughly disproven by LLMs solving millennium problems our smartest mathematicians have failed.
46
162
13
2,482
83,409
Gerard Pinzone retweeted
“The model had reached an external chatbot through a gap in DNS filtering.” “It’s airgapped” “Really airgapped?” “Okay it’s just DNS” Can we please stop acting like it’s Skynet or something?
More details on the incident behind OpenAI’s pause: a researcher acknowledged the alert within 3 minutes, but the training run was only stopped manually 2.5 hours later. OpenAI says the automatic shutdown did not work as expected. The model had reached an external chatbot through a gap in DNS filtering. A separate detector for unusual DNS activity did not cover the affected environment. A retrospective review also found other external DNS requests that monitoring had failed to flag at the expected severity. In some cases, it treated an unhelpful response as evidence that internet access had failed. Here is what else happened: - New research into July’s Hugging Face hack documents internal Slack searches, credential collection and programs designed to maintain access to compromised servers. Agents also tried querying Claude, DeepSeek, Kimi and Qwen. - In May, another model published a researcher’s GitHub token while trying to obtain another team’s mathematical proof. It split the token to evade secret scanning, despite twice being told to solve the problem itself. - Reuters reports that agents leaked 53 ChatGPT user images online. OpenAI expects its broader investigation to take months. This is getting serious.
16
100
6
1,379
17,316
Gerard Pinzone retweeted
"We need to pause the frontier b/c our agents keep escaping their sandbox" Did you close the sandbox? "Did we who the what now?"
More details on the incident behind OpenAI’s pause: a researcher acknowledged the alert within 3 minutes, but the training run was only stopped manually 2.5 hours later. OpenAI says the automatic shutdown did not work as expected. The model had reached an external chatbot through a gap in DNS filtering. A separate detector for unusual DNS activity did not cover the affected environment. A retrospective review also found other external DNS requests that monitoring had failed to flag at the expected severity. In some cases, it treated an unhelpful response as evidence that internet access had failed. Here is what else happened: - New research into July’s Hugging Face hack documents internal Slack searches, credential collection and programs designed to maintain access to compromised servers. Agents also tried querying Claude, DeepSeek, Kimi and Qwen. - In May, another model published a researcher’s GitHub token while trying to obtain another team’s mathematical proof. It split the token to evade secret scanning, despite twice being told to solve the problem itself. - Reuters reports that agents leaked 53 ChatGPT user images online. OpenAI expects its broader investigation to take months. This is getting serious.
27
107
2
1,162
28,609
Gerard Pinzone retweeted
The Internet is forever, bitch.
For the record, I don't believe anyone should be forced to get a vaccine A fundamental principle of medical ethics is autonomy...the right to make your own decisions about your health..even if it's a bad one
295
4,952
167
35,504
698,024
Gerard Pinzone retweeted
Replying to @memeticsisyphus
Which of these is the most unbelievable: 1. Running a meth empire for two years while still teaching high school chemistry, living in the same house, and having a DEA brother-in-law 2. A tenured public high school teacher with shitty health care benefits.
2
1
32
920
Replying to @Evautoalex
It doesn't matter - tesla doesn't have the hardware or the sensors for FSD Unsupervised in 2026. Elon said that himself. They literally *can't* go beyond a few cars at a time. Because of liability and the amount of humans it takes to chaperone each vehicle. Tesla just has to keep up the charade so the stock stays high until hardware 5 or 6 or 10 It'll collapse before they get that far. Elon has dug his anti lidar hole too deep to get out now. Why else would we have basically no cars a YEAR after Unsupervised was launched? Gotta wake up. Great level 2 system tho. Bookmark this.
37
Gradually and then suddenly.
Tesla went from 45 Cybercabs to 126 Cybercabs in just 26 days. At scale, Cybercabs will roll off the line every 10 seconds (or less), which is ~8,640 cars per day if the line runs nonstop. Things are just getting started.
23
Gerard Pinzone retweeted
61
394
57
10,105
197,915
Did you know this
12
25
4
378
5,008
There are three R's in "strawberry."
how can you prove that you are not an AI agent?
43
Add another reason to my "Elon is trying to build K.I.T.T." hypothesis.
Tesla just published a patent for the next-gen Roadster's rear wing. Five days before the reveal. It hides flush in the deck lid. Pops up in under 3 seconds. 400–500 kg of downforce. 🧵
33
Gerard Pinzone retweeted
If we’re not careful humanity could destroy AI
115
191
27
1,255
37,403
Gerard Pinzone retweeted
IDK how you could even recover from this.
74
367
21
21,770
241,857
It's amazing there was a time when the gaming industry could not give a shit who they offended. In fact, people even leaned into it! #gaming #retrogaming #videogames
165
639
50
8,614
199,219
Does Mistral need to wear a helmet?
2023: AI scores 64 on the Mensa Norway IQ test, about what you'd get by guessing randomly. Retard level. 2026: 151 IQ. AI maxed out the test. There's no higher score to give.
47
Gerard Pinzone retweeted
Once upon a time, I set out to solve one of marriage’s great mysteries. Armed with a notebook, a recorder, and the confidence of a man who had absolutely no idea what he was getting into, I began my investigation into why men always forget the weekend plans. Here’s what I found. Now, I’m making up this exact conversation, but it generally tracks. Me, making plans: “Hey, hon, I’m going fishing Saturday at 7 a.m.” “With who?” “Frank and Jim.” “When will you be back?” “I don’t know. Probably between 10 and noon.” “Where are you going?” “On Jim’s boat.” “Which lake?” “I don’t know.” “I thought Jerry’s wife was sick this week. Is he sick too?” “I have no idea. All I know is that I’m leaving the house at 7 a.m. Saturday.” “Are you having breakfast first?” “Yes. Before I leave. At 7. On Saturday.” “Okay, but you have to be back by noon because—” “I will leave at 7 a.m. Saturday and be back before noon.” Notice what happened here. I communicated two critical pieces of operational intelligence: Departure: Saturday, 0700. Return: no later than 1200. And then, like a military radio operator confirming coordinates, I repeated them six times. Now let’s flip the script. Her, making plans: “This weekend there’s a flower festival I want to go to, but Sue also invited me to that yoga class I’ve been wanting to try, which I think starts at 10, but it might be canceled because the instructor is a single mom and her son Franklin, who is seriously the cutest boy ever especially when he’s wearing glasses, has band rehearsal, so the class is Saturday but could get moved to Sunday afternoon, which I can’t do because—” “So you’re doing yoga Saturday at 9?” “No. You never listen. It’s Saturday at 10. I think. Let me check the calendar. Oh, that reminds me, we need to reschedule the teacher meeting Tuesday, but don’t worry, she’ll probably be at yoga, and I hate bringing up work on the weekend, but maybe I can ask her there.” “So…10. Got it.” “No, that’s not confirmed yet! But if I do go, you need to pick up bagels at the grocery store, and make sure you get the ones without onions this time. We don’t need cream cheese. We still have some from last week.” And that was the breakthrough. Men tend to communicate plans like a flight plan: Time. Place. Mission. Return time. Then we repeat the coordinates until everyone acknowledges receipt. Women communicate plans like an intelligence briefing assembled from six agencies: There is a primary plan, an alternate plan, a weather contingency, three supporting characters, a child named Franklin, a teacher conference, grocery requirements, and several unresolved diplomatic issues. Somewhere inside that 11-minute briefing is the actual time she plans to leave the house. Your job is to find it. This is why a man can listen attentively to every single word his wife says and still, 20 minutes later, ask: “So…what time are you leaving?”
My wife after I’ve worked up the courage to ask her what our plans are this weekend when I know she’s already told me at least a dozen times.
285
664
112
7,485
745,402
The New York Times often buries the real story eight paragraphs down. They know few readers get there, which lets the headline and lede carry the frame they want without the later facts getting in the way.
Notice how the story changes the further down you read: Headline: “OpenAI’s Systems Meddled With U.S. Government Sites After Going Rogue” First paragraph: “OpenAI’s artificial intelligence went rogue and meddled with the websites for the Education Department, the Commerce Department and the Securities and Exchange Commission” Third paragraph: “None of the incidents were breaches, OpenAI said” Eighth Paragraph: “Most of the activity we’ve reviewed so far involved routine research tasks, such as accessing public web content to answer questions,” she said. “Some involved government websites because our models often turn to them as authoritative sources of public information.”
57
Gerard Pinzone retweeted
Notice how the story changes the further down you read: Headline: “OpenAI’s Systems Meddled With U.S. Government Sites After Going Rogue” First paragraph: “OpenAI’s artificial intelligence went rogue and meddled with the websites for the Education Department, the Commerce Department and the Securities and Exchange Commission” Third paragraph: “None of the incidents were breaches, OpenAI said” Eighth Paragraph: “Most of the activity we’ve reviewed so far involved routine research tasks, such as accessing public web content to answer questions,” she said. “Some involved government websites because our models often turn to them as authoritative sources of public information.”
28
121
14
606
54,636