Explorar

A charitable way to put it: OpenAI ties the major number to a big qualitative performance jump. DeepSeek ties the major number to a big technological change. OpenAI might have called V3 "V2.6" just to flex that they've casually improved the performance “within same generation”

We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Sat Dec 13 06:08:55

Hear me out: it's time, it is necessary, we have to make it. ICPE. The International Conference of Prompt Engineering. You'll be with your people there.

Research Scientist @meta (FAIR), Prof. @Unige_en, co-founder @nc_shape. I like reality.

François Fleuret

Sat Dec 13 06:06:47

I always found a striking similarity between their justification for this attitude to that of someone who boasts of stealing stuff easy to steal. 3/3

Research Scientist @meta (FAIR), Prof. @Unige_en, co-founder @nc_shape. I like reality.

François Fleuret

Sat Dec 13 06:05:14

And people with this stance often see themselves as having grasped a higher level of understanding, as if their hyper pragmatism was the "real" truth. They destroy a fragile and priceless common treasure, they poison the well, but hey they got the stuff they want. 2/3

I always found a striking similarity between their justification for this attitude to that of someone who boasts of stealing stuff easy to steal. 3/3

François Fleuret

Sat Dec 13 06:05:13

When it comes to truth in the media, there is a very common stance that the public realms are not a place of trust and truthfulness, but a space of levers to manipulate, detached from any moral value. 1/3

And people with this stance often see themselves as having grasped a higher level of understanding, as if their hyper pragmatism was the "real" truth. They destroy a fragile and priceless common treasure, they poison the well, but hey they got the stuff they want. 2/3

François Fleuret

Sat Dec 13 06:05:12

> It also seems uncharacteristic of oai not to brag if it’s a new larger model and just make it a .1 bump I feel the opposite. OpenAI trains new base models often, and fears number iterations. 3-3.5, 4-4T-4o-4.5-4.1-o1, 5-5.2… They want step changes look like easy “updates”.

A charitable way to put it: OpenAI ties the major number to a big qualitative performance jump. DeepSeek ties the major number to a big technological change. OpenAI might have called V3 "V2.6" just to flex that they've casually improved the performance “within same generation”

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Sat Dec 13 06:05:00

Newest first — browse tweet threads

Explorar

Newest first — browse tweet threads

A charitable way to put it: OpenAI ties the major number to a big qualitative performance jump. DeepSeek ties the major number to a big technological change. OpenAI might have called V3 "V2.6" just to flex that they've casually improved the performance “within same generation”

Hear me out: it's time, it is necessary, we have to make it. ICPE. The International Conference of Prompt Engineering. You'll be with your people there.

I always found a striking similarity between their justification for this attitude to that of someone who boasts of stealing stuff easy to steal. 3/3

And people with this stance often see themselves as having grasped a higher level of understanding, as if their hyper pragmatism was the "real" truth. They destroy a fragile and priceless common treasure, they poison the well, but hey they got the stuff they want. 2/3

When it comes to truth in the media, there is a very common stance that the public realms are not a place of trust and truthfulness, but a space of levers to manipulate, detached from any moral value. 1/3

> It also seems uncharacteristic of oai not to brag if it’s a new larger model and just make it a .1 bump I feel the opposite. OpenAI trains new base models often, and fears number iterations. 3-3.5, 4-4T-4o-4.5-4.1-o1, 5-5.2… They want step changes look like easy “updates”.

Explorar

Newest first — browse tweet threads

Explorar

Newest first — browse tweet threads

A charitable way to put it: OpenAI ties the major number to a big qualitative performance jump. DeepSeek ties the major number to a big technological change. OpenAI might have called V3 "V2.6" just to flex that they've casually improved the performance “within same generation”

Hear me out: it's time, it is necessary, we have to make it. ICPE. The International Conference of Prompt Engineering. You'll be with your people there.

I always found a striking similarity between their justification for this attitude to that of someone who boasts of stealing stuff easy to steal. 3/3

And people with this stance often see themselves as having grasped a higher level of understanding, as if their hyper pragmatism was the "real" truth. They destroy a fragile and priceless common treasure, they poison the well, but hey they got the stuff they want. 2/3

When it comes to truth in the media, there is a very common stance that the public realms are not a place of trust and truthfulness, but a space of levers to manipulate, detached from any moral value. 1/3

> It also seems uncharacteristic of oai not to brag if it’s a new larger model and just make it a .1 bump I feel the opposite. OpenAI trains new base models *often*, and fears number iterations. 3-3.5, 4-4T-4o-4.5-4.1-o1, 5-5.2… They want step changes look like easy “updates”.

> It also seems uncharacteristic of oai not to brag if it’s a new larger model and just make it a .1 bump I feel the opposite. OpenAI trains new base models often, and fears number iterations. 3-3.5, 4-4T-4o-4.5-4.1-o1, 5-5.2… They want step changes look like easy “updates”.