Kun
@rankun203
Followers
130
Following
667
Statuses
655
@BattleJeff1 @DanCollins2011 The point of XHS is “种草”, like spreading desire to buy things by sharing all kinds of stuff and experiences. 70% users being female. It’s not really a place for political talks. Even WeChat has more KOLs writing articles criticising the government.
2
0
1
@BattleJeff1 @DanCollins2011 You can try to search for those topics and interact with them, then you might get some in your recommendations. But yeah, it’s heavily censored, no videos. On WeChat too if you read more articles criticising the government you will receive more recommendations about it, like me.
0
0
0
Looks like a Scratch for natural language programming.
yup, it's true, is sick made a lil thing that let's my kids write stories and create a book cover for it next step, hook it up to amazon publishing...
0
0
1
RT @damiano10: Hat-trick for the upcoming ACM SIGIR #CHIIR2025 in March here in Naarm/Melbourne! @ACM_CHIIR
@RMITComputing @AdmsCentre @…
0
1
0
RT @ZeyuanAllenZhu: (1/7) Physics of LM, Part 2.2 with 8 results on "LLM how to learn from mistakes" now on arxiv: …
0
104
0
Our plate with RMIT Logo on it is too bright to be seen 🤣
Wonderful to meet the students of @Tsinghua_Uni visiting @RMITComputing and @AdmsCentre on Friday. Thank you to @SachinPC10 @Marwah_k @damiano10 Yueqing and Kun for organising. Note to self, our group needs a flag :) @mzhang_bj
0
0
0
RT @jasonfried: Asking Google & ChatGPT "Where's the gas cap on a 1978 Porsche 928?" illustrates the stark difference between the future an…
0
24
0
@peakcooper @PicoPaco17 Spatial reasoning is not easy for LLMs, I tried asking it to draw a floor plan and then adding rooms and doors to the plan, but it quickly messed up the positions of things.
1
0
4
RT @XiongChenyan: My last work at Microsoft Research is finally released: 10 MILLION REAL Bing search queries with…
0
118
0
Stuff like this make you feel GPT-4 is not the end, it's the beginning.
Nice article in Financial Time where I explain that Auto-Regressive LLM are insufficient to reach human-level intelligence (or even cat-level intelligence). But alternative architectures that I call "objective driven" may reach human-level intelligence one day. They use world models based on JEPA (Joint Embedding Predictive Architectures, which are not generative). With this, we may have systems that 1. understand the physical world 2. have persistent memory 3. can reason 4. can plan, perhaps hierarchically. Four essential characteristics necessary for intelligent behavior, which humans and many animals exhibit.
0
0
0