deter3 @deter3

develop AI creativity and AI planning , AI + Psychology + cognitive , social media real-time events detection +NLP Shinagawa-ku, Tokyo Joined April 2008

Tweets

4K
Followers

1K
Following

1K
Likes

8K

deter3 @deter3

7 hours ago

Chinese AI researchers are highly practical and consistently deliver results, especially in multi-turn RL research following GRPO.

0 0 2 87 0

deter3 @deter3

3 days ago

LOL , Wang can manage slaves not AGI researchers .

Zephyr @zephyr_z9

3 days ago

LOL , Wang can manage slaves not AGI researchers .

78 135 2K 193K 656

Download Image

0 0 1 296 0

My hypothesis : The empirical and implementation-heavy nature of Large Language Model (LLM) development aligns more closely with Chinese demonstrated strengths in practical, application-driven AI research. Conversely, a tendency towards a more theoretical mindset in the Indian…

0 0 0 103 0

deter3 @deter3

4 days ago

many people believe in "brain in a vat" is just stage 1 > no matter how smart ai is, he's still in a vat > agentic is how we gives ai hands and interaction with the real word. From my personal experience , this might not be the right path to complex tasks , it looks like RL +…

1 0 0 102 0

deter3 @deter3

4 days ago

diffusion model + auto regressive model is the future .

0 0 0 79 0

deter3 @deter3

a week ago

The Litmus Test for a Startup Idea(Inspiration from nano-banana) To know which domain you are in, ask this one question: "Is the final output of my system valuable because it is plausible and creative, or because it is provably correct and verifiable?" If the answer is…

0 0 3 182 2

deter3 @deter3

a week ago

Basically , what LLMs model are doing , or LLMs model generation , is human cognition ensembles/orchestration , or human cognition simulation , since the statistics of linguistics, embedded in LLMs model parameters , are simply a reflection of human collective cognition —…

0 0 1 164 1

deter3 @deter3

a week ago

The Future of Science: The End of "Why" .

0 0 0 65 1

deter3 @deter3

a week ago

Why "The statistics of linguistics embedded in LLM parameters are simply a reflection of human collective cognition": From scraping 1 billion Instagram text posts over 5 years, I discovered clear patterns: "apple" appeared with "milk" in 50% of morning posts initially, but this…

deter3 @deter3

2 weeks ago

0 0 3 398 0

0 0 0 153 0

deter3 @deter3

a week ago

Openai products design is in a huge mess , performance the features fucked up a lot , like i am using o3 , asking myself , is it the o3 i was used to deal with in the 2 months ago ? and deep research is unable to use . Lots of bugs in their website chatgpt products .

0 0 1 109 0

deter3 @deter3

2 weeks ago

The output from Large Language Models, including their so-called "emergent properties," is better understood as a virtual simulation of human collective cognition. There are no true emergent properties involved; it's just the model manipulating the vast number of patterns it…

0 0 0 101 0

deter3 @deter3

2 weeks ago

Most of people get confused on one thing . LLMs can not do reasoning , can not think , because without "prompt" , LLMs model will not generate anything . why no prompt , no generation ? Do you think anything without prompt , then generate nothing is the one who can reasoning or…

0 0 0 102 0

deter3 @deter3

2 weeks ago

The statistics of linguistics, embedded in LLMs model parameters , are simply a reflection of human collective cognition — nothing fancy . You can say it has consciousness or non-consciousness . You can call it has intelligence or not having intelligence . It does not matter…

0 0 3 398 0

deter3 @deter3

2 weeks ago

Don’t just study AI. Make it work inside a business.

TBPN @tbpn

2 weeks ago

Don’t just study AI. Make it work inside a business.

69 170 2K 372K 1K

Download Video

0 0 0 154 0

deter3 @deter3

3 weeks ago

Using AI, I created an industry breakthrough by simply describing what I wanted. AI handled the detailed planning and coding for what's normally a 3-6 month project.

0 0 2 113 0

deter3 @deter3

3 weeks ago

Engineering shouldn't be fully autonomous. If design, development, deployment, and operations are all automated, what role is left for human engineers? This autonomous approach to coding and system building eliminates the need for engineering expertise entirely. Where the…

0 0 0 101 0

deter3 @deter3

4 weeks ago

GPT-5 demonstrates that you should never predict the future in the AI industry. Progress isn't smooth or predictable, and major setbacks can occur suddenly, similar to the AI winter of the 1960s. Those who make predictions and promise what will happen are either lying or…

1 0 1 170 0

deter3 @deter3

4 weeks ago

In my view, DSPy repeats LangChain’s pitfalls: heavy abstractions over simple workflows, limited proof of value on complex tasks, and messaging (“Better Together,” outperforming GRPO, CPU RL) that outpaces demonstrated results. Since RL drives behavior through data and reward…

0 0 0 112 0

deter3 @deter3

4 weeks ago

At a philosophical level, software engineering conflicts with LLMs. FSD V12 is A very good example , after end to end learning , FSD v12 need much less code while v11 needs huge amount of code . why so ? Fundamentally, software engineering is a human-led attempt to automate…

deter3 @deter3

4 weeks ago

2 1 4 717 1

1 0 4 203 1

deter3 @deter3

4 weeks ago

LLMs require entirely new mental models, not retrofitted software engineering concepts. Andrej Karpathy is till holding a software engineering mindset Many AI researchers are stuck in the wrong paradigm They approach LLMs with a software/automation mindset - thinking…