news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

hi, hi, although not at the same time, but at the same place, i’m here at the yunqi conference again this year!

it must be said that this year's road to yunqi town was once again plagued by the annual classic traffic jam. the forum opened at 9:30 in the morning, but we were stuck in traffic until 9:15 before we could get in.

although we were not the first to arrive, we were still quite ahead of the other teams that were stuck in traffic behind us. after all, we in hangzhou still have some home advantage.

there were already a lot of people on our opening day, and there were even more people at the closing ceremony today. the short holiday has just passed, and many children came to the scene during the weekend, so it was really lively.

i don’t know if my friends in hangzhou have been to the event. friends in other regions, don’t panic. let’s quickly show you what’s new about this yunqi conference!

first of all, just talking about the main forum of this yunqi conference, there are many highlights, becausethe specifications this time are really high。

hangzhoumayor, deputy secretary of the municipal party committeewhen the host, alichairman wu yongmingandacademician wang jianalso at the scene, big names such as he xiaopeng and kimi ceo also participated in the roundtable discussion.

wu yongming also made some golden sentences in this speech, such as "the future of ai is not on mobile phone screens, but in taking over the digital world and changing the physical world."

at the roundtable discussion, the ceo of jieyuexingchen, yang zhilin of kimi, and the ceo of shengshu technology also expressed sharp opinions. for example, many people believe that ai seems to have encountered a bottleneck in this year, and the speed seems to have slowed down and is not as explosive as before.

but as practitioners, they feel thati didn't thinkAI the development speed is slow, but it can be said that there has never been such a fast technological update and iteration in human history.(when we interviewed zhipu ceo before,his point of view is also the ceiling that has not yet been seen ）

effective accelerationisme/ acc), silicon valley’s new thinking under the development of ai

in addition, they also talked about openai. they also spoke highly of o1, which is not like what some people think: "openai is dead and o1 is just an engineering optimization."

the reason is that in the current large model field, although the number of parameters and data have been rolled up before, the only effective method is still the scaling law (using a large amount of data, computing power and general architecture to achieve intelligent emergence through "brute force solution"). openai o1 found that volume reinforcement learning also complies with the scaling law, which means that as long as we can continue to scale up (larger scale) and find more paths that can support scaling up, it will be very valuable.

in layman's terms, the success of previous large models mainly relied on piling up parameters and data, but now openai o1 has created a new "expansion route" for proof reasoning.this means that we can continue to expand the scale and find more effective stacking methods, and new things will continue to emerge in this field.

but we think that this may also be related to the fact that most of these founders have academic backgrounds. as young intellectuals, they are all very idealistic people and firmly believe in the agi route, so they are very confident in the future of ai.

however, this conference was not all about speeches. in addition to the main forum, the conference also held three exhibition areas, invited a number of ai and cloud computing manufacturers, and manyfamiliar faces that no one expectedhe's here too. i'll keep it a secret for now, we'll talk about this later.

there are three venues in total this time, the ai ​​pavilion, the cloud computing pavilion and the ai ​​application pavilion. the application pavilion is located in the corner of the park, and it is a bit far to walk to, about eight or nine hundred meters, and it is also the most attractive of the three.

although you may have seen some of these things before on the internet or at other exhibitions, there are also some interesting gadgets this time. they are quite interesting when you get close to them. let me show you some of them below.

for example, as soon as you enter ai hall no. 1, you will see a whole wall of action figures, all of which are 3d printed!

we even played with it for a few times, and you know what, it feels really authentic to hold, and the workmanship is quite fine.

although this thing was actually launched at the end of last year, vast's ai 3d large model tripo is indeed very capable. it can directly convert 3d files by feeding in two-dimensional pictures, or generate them yourself with prompt words, or generate them yourself with pictures + prompt words, or fine-tune them.

after communicating with the on-site staff, we learned that the generated 3d file can also be imported into 3d software for adjustment. to be honest, this may be much better than scanning and modeling on mobile phones.