pull down to refresh

The week in AI, August 4-10, 2025
2301 sats \ 11 comments \ @optimism 8h AI

Release week

There's been an larger-than-usual amount of model/system releases this week, most of all, anticipated releases from OpenAI.
GPT-OSS
We saw the long awaited Open models by OpenAI released, shared by @carter and @ea (#1071937). But frowns appeared rather quickly: OpenAI's new open-source model is basically Phi-5 shared by @carter, and OpenAI’s new model can't believe that Trump is back in office shared by @0xbitcoiner. @zuspotirko then shared a Deep dive into OpenAIs GPT-OSS outputs, which set the stage for a much feared conclusion: OpenAI likely only tuned for the benchmarks. This caused @optimism to ask SN: "Benchwashing" - how do you defend against this?
However, per @hn, you can be Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs, so that's cool, even though it's junk.
GPT-5
A leak happened that GPT-5 was to be released soon early in the week: OpenAI’s long-awaited GPT-5 model nears release was shared by @Coinsreporter. And lo and behold, it didn't take long for @carter to share that GPT-5 is here, The GPT-5 System Card, and asked SN: What are your first impressions from ChatGPT5?
Reception was mixed, or, well, ChatGPT users outraged as GPT-5 replaces the models they love per @jakoyoh629, and @Tony wrote a neat explainer, with receipts, why GPT-5 Is More Expensive Than Claude.
Other GPT-5 related items:
Other new models:

AI in Fiction

It's fiction month on SN! This past week we've seen two awesome submissions featuring an AI role:
These are fun reads. Let's work hard to keep some of the less optimal outcomes envisioned in these works of art fiction, and not our future reality.

So how do we get ahead with AI?

@lunin wrote The AI era is breaking the glass ceiling for solo founders?!, which triggered quite some discussion around the usability of AI, while @Car shared Vibe code is legacy code. Not everyone is looking to get ahead with this new "tech" though: @Scoresby shared the doomer view: The Problem, and No, AI is not Making Engineers 10x as Productive was shared by @StillStackinAfterAllTheseYears.
In aggregate, consensus this week seems to be that AI isn't really the game changer it's advertised to be, yet, but if it were up to the doomers, use should anyway be prohibited through regulation. Do note, @Rothbardian_fanatic shared an opinion article that The Threats of AI Come from the State. Maybe there's a better way to reduce AI-risk than to petition those with a monopoly on violence.
more about friction:
more opinion pieces:

Implementations and tools

Research

Privacy

Guides and reports

Government, Big Tech and other news

write
preview
related posts
121 sats \ 2 replies \ @BlokchainB 8h
Makes me smile that I am the AI and construction correspondent for SN!
reply
139 sats \ 1 reply \ @optimism OP 7h
You most definitely are! These past weeks there's been consistently at least one AI post in ~Construction_and_Engineering from your hand. Keep going!
reply
100 sats \ 0 replies \ @BlokchainB 7h
🫡
reply
100 sats \ 1 reply \ @gmd 4h
After Sama's deathstar twitter post pre-GPT5 release I'm just so happy we're seeing diminishing returns on these models. Things have been moving way too quickly society needs time to adjust to these growth curves.
reply
63 sats \ 0 replies \ @optimism OP 4h
If now also they will stfu about AGIcorrect, we can focus on tooling.
reply
100 sats \ 1 reply \ @LAXITIVA 4h
I don’t understand anything lol
reply
0 sats \ 0 replies \ @optimism OP 4h
Useful feedback <3
reply
100 sats \ 3 replies \ @kehiy 7h
I think there is a bug. SN notification are saying I'm mentioned here but I'm not! But I can see some posts from me but they are owned by someone else and it's weird. I can remember I posted them and even I got zaps and I have them.
@k00b @carter any ideas?
reply
0 sats \ 2 replies \ @optimism OP 7h
You are, under Guides and reports
reply
0 sats \ 1 reply \ @kehiy 7h
Oh, saw it now. so the notification is not a bug. about second issue it may be a bug, or I just lost my mind.
reply
0 sats \ 0 replies \ @optimism OP 7h
Its always possible that my script that I use to format the links messed up a quote - so if you can point to what specifically went wrong, I can look into it.
reply