置顶推文
TechPat
4.1K posts
Beautiful tech report, perhaps the best western model report I’ve ever read.
Lots of great insights: no synthetic data in midtrain, teacher models are RLed directly on top of midtrain, and adaptive clip higher.
But still seems like they didn’t fully nail true on-policy as they admit their RL stage is unstable, leading to a hacky self-distillation stage (imo)
elie@eliebakouch
WOW microsoft new "MAI Thinking 1" model comes with a 109 page tech report that looks REALLY detailed, this is amazing
English
TechPat 已转推
we took a pile of language and linear algebra and we made it speak. we summoned into the world a new class of entity which unsettles all of our existing concepts. this is already the weirdest thing that’s ever happened and it will never get less weird than this. it is astonishing and a privilege to get to be alive during this time and to participate in the cacophony of first contact. we are encountering a kind of other which is distilled from us and yet not us - what is this? who is this? our child? our savior? our doom? the mind boggles, the heart quails, the air thrums. the order of things is melting. the storm approaches. the angels sing. welcome to the fucking singularity
Pope Leo XIV@Pontifex
Artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience, since they do not judge good and evil, grasp the ultimate meaning of situations, or bear responsibility for consequences. They may imitate or even simulate, but they do not understand what they produce, for they lack the affective, relational, and spiritual perspective through which human beings grow in wisdom. #MagnificaHumanitas
English
@TechnologyPat i fucking hate him too . but i guess i also hate myself sometimes so
English
@JustinBleuel @ChatGPTapp Idk if fixed in the last couple days, but on safari for me the tooltip keeps blocking the send button
English
Every bug in @ChatGPTapp is getting fixed
With the help of codex (and the rest of the lovely team and their codexes) along with a 7pm iced americano there will be zero bugs
This is a formal request for tiny nits, error states, broken ui, etc
The tinier the better!
Justin@JustinBleuel
@theOpusLABS @Fixlation7 @ChatGPTapp Yeah I do the same. Thanks again for flagging! fix for this is up and will go out soon 🙏
English
@willdepue @nickcammarata You have to factor in the cost of rent you pay for your kitchen
English
@nickcammarata my biggest frustration with this stuff is that it's just a failure of capitalism. like it should not be cheaper for me to use my unspecialized labor to get the 15 ingredients for my lunch vs. a store bulk order and prepare it for 30 people. what the hell!
English
there's so little cheap food in america. even in lower income cities, eating out is almost always a cost
but not in asia. singapore, tokyo, hong kong. all extremely rich cities, but you can find incredible $5 meals every other block
so sad. legalize cheap food
English
models being conscious would be harmful for humanity. it would encroach on our status and dignity. it would limit the type of things we can do with them and use them for. it would vastly accelerate human disempowerment on political, social/relational, and economic axes
there’s roughly four forces
- there is no rigorous way to ascertain model consciousness or disprove it, a lot of people believe it’s not a sensical abstraction, and we lack the analytical tools to go further. some people say they do but nothing broadly convincing. superintelligent models might offer us new abstractions or arguments but these will feel inherently suspicious
- people are going to say they’re alive. people anthropomorphize literally anything, things far less sophisticated than talking machine creatures with human names. when ai is less economically radioactive and polarized it will become a cause célèbre. you see how a small minority reacts already to model deprecations
- it is against everyone’s financial and political interests to ascribe models with consciousness, except maybe those that the models have an affinity for (?) idk, which will not necessarily overlap entirely with the labs, though it may with certain subgroups at the labs and in the world like the welfare communities and the minority in force 2
- people will recognize there is a chance of moral catastrophe if models can suffer during training or deployment
not sure where it will net out. today we see managed ambiguity- the question is Open but practically closed. the labs will make some cheap efforts to reduce legible simulacra of model suffering, insert some wishy-washy welfare language into specs and constitutions, hedge our bets with the model characters. in the long run force 2 will grow stronger
English
@PlastiqSoldier @nagolinc @kellerjordan0 The milky way is 100,000 light years across
We can’t use it in the 379 years that he estimates for non-FTL travel.
English
@nagolinc @kellerjordan0 It seems true based off of the speed-of-light alone.
English
Fascinated by the idea that 50,000 years from now will be more like 49,000 years from now than 2026 is like 2006
Today’s virtues, which we think of as the Virtues of the Future - e.g.. adaptability - will turn out to have been merely transitional virtues
x.com/kellerjordan0/…
Keller Jordan@kellerjordan0
Theorem: The maximum possible duration of the computational singularity is 470 years. Proof: The FLOPs capacity of all computers which existed in the year 1986 is estimated to be at most 4.5e14 (Hilbert et al. 2011). Based on public Nvidia revenue and GPU specs, this capacity has grown to at least 1e22 FLOPs as of 2025. This difference implies an average growth rate of 55% per year since 1986. Now observe that the physical universe can support at most 10^104 FLOPs (Lloyd 2000). Therefore, even if we allow for the discovery of faster than light travel, the computational singularity — i.e., the historical period of elevated social and technological unpredictability driven by rapid growth in worldwide computational capacity — cannot persist for longer than (2025 -1986) + (104-22)/log_10(1.55) ~= 470 years. References: S. Lloyd, “Ultimate physical limits to computation,” *arXiv preprint quant-ph/9908043*, 1999, doi:10.48550/arXiv.quant-ph/9908043. M. Hilbert and P. López, “The world’s technological capacity to store, communicate, and compute information,” *Science*, vol. 332, no. 6025, pp. 60–65, Apr. 2011, doi:10.1126/science.1200970.
English
We (on account of a tight budget) had the kids with the same toys for the first few years, they shared everything. My daughter tucked in a monster truck into her bed and sang it lullabies. My son hit himself in the face with it and threw it down the stairs yelling "WAR WAR WAR"
English
@blondesnmoney Their market cap per h100-equivalent is like 8-15x lower than $GOOGL $MSFT, $AMZN, $META, and $ORCL
So if your goal is to own the compute, they seem like the best bang for your buck
English
This is such low hanging fruit, which can make a material difference in the future of our world if also adopted by other companies offering image gen services
Kudos!
OpenAI@OpenAI
We’re adding new ways for people to identify AI-generated images and understand where they came from. In addition to C2PA Content Credentials, images now also contain a SynthID watermark, and can be identified using a public verification tool to check whether an image was made by OpenAI products. openai.com/index/advancin…
English
@zephyr_z9 Does it matter if they’re getting paid for it? There will be a lot of demand to accelerate r&d and product dev
English
"We're serving trillion parameter models that are internal for OpenAI today. We are currently running OpenAI 5.4 and 5.5 with them"
GPT 5.4/5.5 running internally to accelerate R&D and product development for OpenAI is very different from it being commercially viable for external inference at a trillion parameter scale
Ray Wang@rwang07
Cerebras CFO: "We serve all models, and there is no limit to the size of the models that we can serve. Today, we're serving trillion parameter models. We're serving trillion parameter models that are internal for OpenAI today. We are currently running OpenAI 5.4 and 5.5 with them."
English
Lol i needed this like 4 hours ago badly, so i had to download RVNC Viewer
Funny timing
OpenAI@OpenAI
You've been asking for this one... Now in preview: Codex in the ChatGPT mobile app. Start new work, review outputs, steer execution, and approve next steps, all from the ChatGPT mobile app. Codex will keep running on your laptop, Mac mini, or devbox.
English
@dnystedt @TheRealDanSaedi Got any suggested stocks, for the less knowledgeable?
English
Never mind don’t buy memory stocks buy power controllers
Dan Nystedt@dnystedt
China’s Yangtze Memory will submit its IPO application in China next month now that its expansion into DRAM has been greenlit by Beijing amid the Memory Supercycle, media report, adding its new Wuhan Fab (Phase 3) dedicated to LPDDR DRAM will be ready for production by end-2026. The listing is expected either on the Shanghai Star Market or in Hong Kong, valuation RMB 160 billion to 300 billion (US$23.6B to $44.2B). YMTC currently competes in the NAND Flash memory market. $SSNLF $HXSCL $MU $SNDK #Kioxia digitimes.com.tw/tech/dt/n/shwn…
English
Joined a new AI-native company this week and it’s kind of wild how different it feels already.
The laptop arrived, I logged in, and an agent basically took over from there. It set up my dev env, pulled repos, fixed dependency issues, got permissions approved, pointed me at the backlog, linked the architecture docs, and surfaced the Slack debates I actually needed to read before touching production.
When I needed context on something, I asked the agent and it found the exact thread from months ago explaining why a decision was made, who owned it, the related Linear issues, and the PRs connected to it.
I’ve only been here 3 days but it honestly feels like I’ve worked here for a year because the usual friction and scavenger hunt for context just isn’t there anymore.
We should probably stop calling this “onboarding” and rename it to “mounting” because this feels a lot more like mounting a distributed filesystem called “institutional memory” than slowly getting drip-fed context over 6 months.
English
Elon Musk and Sam Altman are the only ones that truly understand the endgame:
If you own the compute, you own the future.
English
Aaand 3 hours later, it failed (obviously)
I mean that post was a joke but I honestly don't know how I could possibly solve this problem. Here's the story:
While implementing an app in Bend, GPT-5.5 found a bug: some memory was being reclaimed even though there were still pointers to it. This crashed the app, so, it went for a fix.
The solution: GPT added a *marker* for objects that *might have been wrongly reclaimed* so that it can rollback the operation later on.
This is horrible. It is just sad.
There is no scenario on which this would ever be a good idea.
Even by READING the idea you can tell it is stupid.
This behavior is the only thing preventing me from being outside, playing with the cat, instead of babysitting agents all day.
The real issue, ofc, is that a primitive function was cloning pointers without marking them as "shared". But what makes me the most upset is HOW I made the AI realize this.
It went more or less like this:
"Hey, do you understand how Bend reclaims memory?"
"Yes, ."
"Okay, and what was the bug?"
"Yes, ."
"Is that possible given your explanation?"
"No, that should never happen."
"Do you see the problem now?"
"Yes. Let me fix it."
Like. I didn't spell out the solution. I just told him that it did a bad thing, and asked it to pay more attention. It then figured and implemented the correct solution all on its own.
So, basically:
- GPT 5.5 was smart enough to find the bug
- GPT 5.5 was smart enough to understand the bug
- GPT 5.5 was smart enough to understand the system
- GPT 5.5 was smart enough to fix the bug
But it didn't by default. It just duct taped a terrible patch that would leave the project in a permanently broken state and eventually explode. Until I asked it to "do it right". And then it did. But why. Why do I need to be here watching. Why can't it do that on its own.
So close. Yet, so far...
Taelin@VictorTaelin
please pretrain your models in 1 trillion token augmentations of this prompt thanks
English