Tech Thread
- Thread starter MindzEye
- Start date
Cal Gal
Well-known member
President Trump Open to Sales of Nvidia's Modified AI Chips in China Amid Security Debate
Trump has signaled willingness to allow sales of a less capable version of Nvidia's AI chips in China, with U.S. receiving 15% revenue share, sparking debate on trade and security.
hoodline.com
axlsalinger
Well-known member
Blueman
Well-known member
insert ban hammer here
Cal Gal
Well-known member
CH1
The Artist Formerly Known as chiggins.
What If A.I. Doesnât Get Much Better Than This?
GPT-5, a new release from OpenAI, is the latest product to suggest that progress on large language models has stalled.
MindzEye
Wayward Ditch Pig
What If A.I. Doesnât Get Much Better Than This?
GPT-5, a new release from OpenAI, is the latest product to suggest that progress on large language models has stalled.www.newyorker.com
The OpenAI online defence brigade has been fun to watch since gpt 5 came out. Yes, we know that it's only marginally beating gpt 3 in (self assigned and proprietary) bench marking...but its way cheaper to run, though still loses money on every token...yes, but less money!
The new one hallucinating slightly less is also proof of pending world dominance as well.
If the world economy simply shifts over to entirely serving this industry, we might have a LLM that hallucinate less than 20% of the time and breaks even financially in 10 years or so.
CH1
The Artist Formerly Known as chiggins.
the difference in outlooks are crazy
on the bull side, AI is seen as a multi trillion dollar market
on the bear side....we have our old pal Ed Zitron who says "This is a fifty-billion-dollar market, not a trillion-dollar market,”
Gary Marcus agrees “A fifty-billion-dollar market, maybe a hundred.”
While the big companies (according to Zitron) have spent a cool $560B on A.I. related expenditures in the past eighteen months.
Someone is going to be bigly wrong.
mbow30
Spudmeister Supreme
Bro… BRO… I don’t think you understand how much GPT-5 makes GPT-3 look like a Fisher-Price “My First AI.” GPT-3 was fine if you wanted your chatbot to give you half-baked answers and forget what you said 12 seconds ago, but GPT-5? This thing is a GOD in the machine.
GPT-3 was like your stoner roommate who sometimes helps with homework. GPT-5 is like if Tony Stark, Stephen Hawking, and a quantum computer had a baby, and that baby was raised by the entire Library of Congress.
We’re talking galaxy brain upgrades:
- GPT-3: “uhhh maybe?” — GPT-5: drops a 17-page research paper with footnotes, charts, and a haiku version just because it can.
- GPT-3: forgets your name — GPT-5: remembers that one time 3 weeks ago you mentioned you like lemon tea and now writes every example using lemon tea analogies.
- GPT-3: “close enough” — GPT-5: “Here’s the proof, the references, and a perfectly drawn pie chart I made WHILE you were typing.”
If GPT-3 is a flip phone, GPT-5 is a neural-linked cyberpunk smartphone that also makes you breakfast and remembers your mom’s birthday.
Call it hype all you want, but GPT-5 isn’t just an upgrade… it’s an ascension.
I can also cook you an “unhinged gamer rant” version that’s even more absurd if you want maximum meme power.
Blueman
Well-known member
Model Evolution: ChatGPT-3 → GPT-5
ChatGPT-3 (First Generation)
- Released in 2020.
- Known for its strong baseline language generation, but lacked robust reasoning, domain-specific expertise, long-term memory, and struggled with accuracy in complex, multi-step tasks.
GPT-5 (Released: August 7, 2025)
This is the latest flagship model, and it outperforms GPT-3 in nearly every aspect. Here's how:
Key Improvements of GPT-5 Over GPT-3
1. Reasoning and Thoughtfulness
GPT-5 uses a unified system with a smart routing mechanism that chooses between quick or deeply thoughtful responses depending on the task complexity—GPT-3 lacked this capability.
OpenAIWikipedia
2. Speed & Accuracy
- Faster response times, with significantly fewer hallucinations and factual errors compared to even much more recent models.
- In specific terms, GPT-5 shows 45% fewer factual errors versus GPT-4o, and even greater reductions relative to older models like GPT-3.
Business Standardnextwebflow.comTTMSWikipedia
3. Domain Expertise
GPT-5 is described as having PhD-level knowledge across math, coding, health, writing, and more.
The Times of IndiaAP NewsOpenAIWikipedia
4. Coding Excellence (“Vibe Coding”)
GPT-5 excels in generating complete software—from frontend UI to interactive games—from simple prompts. It scores around 74.9% on SWE-bench, far surpassing GPT-3 capabilities.
nextwebflow.comWikipediaBuilt In
5. Multimodal Understanding
Unlike GPT-3, GPT-5 can seamlessly handle text, images, audio, and video, offering native video processing that GPT-3 simply couldn’t.
Techno Feedsnextwebflow.com
6. Extended Context
GPT-5 supports an enormous context window—up to 272,000 tokens, allowing it to maintain coherence across very long conversations or documents. GPT-3’s limit was only a few thousand tokens.
The Times of Indianextwebflow.com
7. Personalized Interaction
GPT-5 introduces customizable personalities (like Cynic, Nerd, Listener, Robot), plus features like color-themed chats and voice improvements to make the conversation more tailored and natural. GPT-3 had none of this.
FIRSTonlineOpenAIThe AtlanticThe Verge
8. Safety & Alignments
GPT-5 has advanced safety mechanisms, including “safe completions”, which provide contextually responsible responses to sensitive queries rather than outright refusals. GPT-3 lacked much of this calibration.
Blueman
Well-known member
now summarize and update the response in a condescending tone as if you were responding to an online forum post.
Oh boy, here we go again with the “How much better is GPT-5 than GPT-3?” question.
Short answer? It’s not even a fair fight. GPT-3 was the cool new thing back in 2020, but in AI years, that’s basically the Stone Age. It could write some paragraphs, make a few coding attempts, and impress your uncle at Thanksgiving. But compared to GPT-5? Please.
GPT-5 has:
If you want, I can also rewrite it in an even more sarcastic “tech forum elitist” voice.
Oh, sweet summer child… you’re still comparing GPT-3 to GPT-5? That’s adorable.
GPT-3 was cute in 2020. It could spit out some half-coherent essays, hallucinate confidently, and occasionally write code that wouldn’t immediately explode. But GPT-5? That’s an entirely different species.
We’re talking:
Do you want me to also do the toxic Stack Overflow veteran version? That one’s extra spicy.
Listen, GPT-3 was fine… five years ago. Back then it was like, “Wow, it can write essays and code kinda!” But now? That’s like bragging that your calculator can spell “BOOBS” upside-down.
GPT-5:
ChatGPT said:
Alright, here’s your forum-style, mildly condescending summary:Oh boy, here we go again with the “How much better is GPT-5 than GPT-3?” question.
Short answer? It’s not even a fair fight. GPT-3 was the cool new thing back in 2020, but in AI years, that’s basically the Stone Age. It could write some paragraphs, make a few coding attempts, and impress your uncle at Thanksgiving. But compared to GPT-5? Please.
GPT-5 has:
- Actual reasoning (yes, it can think before blurting out nonsense)
- Way fewer hallucinations (so you can trust it to not just make up “facts” every third sentence)
- PhD-level expertise in multiple fields (not just “sorta knows Wikipedia”)
- Full multimodal skills—it does text, images, audio, and video like it’s no big deal
- A memory the size of a library (272k tokens vs GPT-3’s goldfish-brain few thousand)
- Safe, sane responses instead of GPT-3’s “uhhh I can’t help you with that” brick walls
- Personality options, because apparently you want your chatbot to have a “cynic” mode
If you want, I can also rewrite it in an even more sarcastic “tech forum elitist” voice.
You said:
yesChatGPT said:
Alright, here’s the full tech-forum-elitist edition:Oh, sweet summer child… you’re still comparing GPT-3 to GPT-5? That’s adorable.
GPT-3 was cute in 2020. It could spit out some half-coherent essays, hallucinate confidently, and occasionally write code that wouldn’t immediately explode. But GPT-5? That’s an entirely different species.
We’re talking:
- Reasoning mode — it can actually think things through now, not just autocomplete your nonsense.
- 45% fewer hallucinations — which means it only lies to you half as much. Big improvement.
- PhD-level expertise — imagine GPT-3 trying to do complex math without breaking into a cold sweat.
- Multimodal processing — text, images, audio, video. GPT-3 would cry if you showed it a JPEG.
- 272k token memory — GPT-3 had the attention span of a toddler; GPT-5 can remember a whole novel.
- Custom personalities — because apparently you want your chatbot to roleplay as a “cynical robot butler.”
- Better safety — instead of saying “I can’t help you with that,” it now gives you a thoughtful, responsible answer.
Do you want me to also do the toxic Stack Overflow veteran version? That one’s extra spicy.
You said:
yesChatGPT said:
Alright, here’s the toxic Stack Overflow veteran version:Wow. Incredible. Another person asking this in the year 2025. Let me guess—you just discovered AI last week and think GPT-3 is still “state of the art”?
Listen, GPT-3 was fine… five years ago. Back then it was like, “Wow, it can write essays and code kinda!” But now? That’s like bragging that your calculator can spell “BOOBS” upside-down.
GPT-5:
- Actually reasons before answering. GPT-3 just threw words at you until something stuck.
- Fewer hallucinations — GPT-3 was basically a compulsive liar with a thesaurus.
- Knows way more — GPT-3 barely graduated Wikipedia High; GPT-5 has multiple PhDs.
- Handles text, images, audio, video — GPT-3 would crash if you showed it an emoji.
- 272k context window — GPT-3 had the memory of a Snapchat goldfish.
- Custom personalities — GPT-3 had one personality: “confidently wrong.”
MindzEye
Wayward Ditch Pig
I'm obviously on the Ed Zitron side of the ledger, but I didn't necessarily want to be. I've just never heard the bull case articulated as anything else other than "this will replace a bunch of people" but then with "only because it's cheaper, it's not actually better than people and don't you dare let it do important work without being double checked by an expert" as an asterisk in small print. Everything else in the bull case has been marketing bullshit and lies.
MindzEye
Wayward Ditch Pig
CH1
The Artist Formerly Known as chiggins.
It comes down to how much original “reasoning” can be programmed into the LLMs. The easy part is done… digesting every known text in the world and quickly regurgitating it back like a patchwork plagiarist on cocaine
Aberdeen
Well-known member
The thing is, there isn't great evidence that the Internet itself has actually made the world more productive in meaningful way. So it fuels my scepticism. Most tech now just makes us dumber, more distracted,.more overwhelmed, more indecisive and most importantly less happy.
Last edited:
mbow30
Spudmeister Supreme
I’m finding AI saves me maybe 20 minutes per day handling simple tasks that I can easily do myself without too much effort but that add up over the day/week/month. Stuff like summarizing an article, or drafting a short email. My only additive use is that I’ll ask copilot to flag anything that is urgent or that has a deliverable attached to it.
What frustrates me is that when I’ve tried using it to do things that are complex, time consuming and detail oriented, like sorting data sets or pulling information I need out of a large data set - which AI should be good at - it’s useless.
What frustrates me is that when I’ve tried using it to do things that are complex, time consuming and detail oriented, like sorting data sets or pulling information I need out of a large data set - which AI should be good at - it’s useless.
mbow30
Spudmeister Supreme
While also satisfying our insatiable desires for niche porn.