toriset

383 posts

toriset banner
toriset

toriset

@torisetxd

The real toriset @toriset on discord

China Katılım Nisan 2024
43 Takip Edilen11 Takipçiler
ratte 🟪
ratte 🟪@rattecs·
what do these three colours remind you of??
ratte 🟪 tweet media
English
213
13
1.6K
219.9K
Endre Stølsvik
Endre Stølsvik@stolsvik·
@teortaxesTex They have made a new thing on OpenRouter; Service Tiers. So there is a «Slow».
Endre Stølsvik tweet media
English
1
0
9
7.5K
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex·
Google should do a reverse-Anthropic and add Gemini 3.5-SLOW 3-Flash was fast enough, and its cost made it attractive this thing is maybe better in some ways but that's not worth the 3x price hike
BURKOV@burkov

I was super excited reading this and was just about to ask Codex to replace, in ChapterPal, all calls to Gemini 3 Flash with Gemini 3.5 Flash, but then I checked the price and decided to stick to Gemini 3 Flash. The new Flash is priced almost like 3.1 Pro, so I'm expecting that the new 3.5 Pro will be priced even higher. If (or rather when) Google phases out 3 Flash, I'm not sure I'll know what to replace it with for visual reasoning about documents. Gemma 4 31B is almost as good as 3 Flash, but it's only available via OpenRouter where, so far, it's unreliable and slow. Any suggestions?

English
4
1
61
6.3K
alex
alex@alexworkmode·
testing something
English
42
1
156
6.6K
toriset
toriset@torisetxd·
@LukeParkerDev check back in a month please, too expensive for a flash-named model, should've used a different name if you're gonna raise the price 3x
English
1
0
1
477
Luke Parker
Luke Parker@LukeParkerDev·
Are you guys finally with me about Google, or do I need to check back in a month?
English
31
2
115
17.2K
toriset
toriset@torisetxd·
@sama wouldnt the plan behind this be rather to sell the spare compute you guys have? i assume the sheer amount of expansion leads to a lot of being wasted if training isnt actively running
English
0
0
1
779
Sam Altman
Sam Altman@sama·
we will offer this until we sell out of our current allocation for this program. (we will make sure to leave enough capacity for ChatGPT, Codex, etc.) we plan to offer it again in the future; our intention remains to build as much compute as fast as we can.
English
101
15
939
155.8K
Sam Altman
Sam Altman@sama·
customers are increasingly asking us for certainty on capacity. as models get better, we expect that the world will be capacity-constrained for some time. we are offering discounted tokens for 1-3 year commits. (it also helps us plan, so hopefully a big win-win.)
OpenAI@OpenAI

Introducing OpenAI Guaranteed Capacity: a new offering that enables customers to guarantee long-term access to OpenAI compute. We’ve made long-term investments in infrastructure, partnerships, and capacity planning to help customers scale reliably. Now, Guaranteed Capacity helps customers plan ahead for critical workloads in a compute-constrained world. openai.com/guaranteed-cap…

English
656
225
5.4K
1.1M
toriset
toriset@torisetxd·
@draecomino groq already does this when its not under load, and also its 32B active parameters, not nearly as impressive as actually a 1T param dense model.
English
0
0
0
226
Adam Holter
Adam Holter@AdamHoltererer·
Gemini 3.5 Flash Frontend Test: Without Skill vs. With Skill
Adam Holter tweet mediaAdam Holter tweet media
English
22
3
294
43.5K
Cerebras
Cerebras@cerebras·
Cerebras is now running Kimi K2.6 – a trillion parameter model – in enterprise trials. At ~1,000 tokens/s, this is the fastest frontier model performance ever measured by Artificial Analysis @ArtificialAnlys.
Cerebras tweet media
English
166
309
4.2K
772.8K
˗ˏˋ Jesse Hanley ˎˊ˗
˗ˏˋ Jesse Hanley ˎˊ˗@jessethanley·
something I keep thinking about: @cursor_ai I hit $1-2k/mo in overages on the ultra plan + have a @claudeai max $200 plan that limits out + looking at a @OpenAI codex plan too all three are heavily subsidised and if i was to use the API I would likely burn +$5k/mo in tokens. how long can this gravy chain be sustained?
English
18
0
141
48.5K
toriset
toriset@torisetxd·
@0xSero 700M in a day is crazy, i barely get that in 3 weeks
English
0
0
0
115
0xSero
0xSero@0xSero·
8.8B this month on Codex. 6.3B in Droid 1.5B in Claude Weak
0xSero tweet media
English
14
1
93
6.4K
toriset
toriset@torisetxd·
@pigeon__s @teortaxesTex that makes no sense, pro is their leading model, that would be the direct competitor to 5.5/5.6, and a competitor to openai's Pro models is DeepThink.
English
1
0
3
94
ρ:ɡeσn
ρ:ɡeσn@pigeon__s·
@teortaxesTex 3.5 Flash kinda does have to beat 5.5 even though itll be cheaper because oAI are just gonna release 5.6 probably in like 2 weeks and Google takes way longer than oAI to ship so they need a 5.6 competitior not a 5.5 competitor so Flash needs to beat oAI's previous best
English
3
0
15
1.3K
toriset retweetledi
Elaina
Elaina@Elaina43114880·
@torisetxd Yes, it was already relatively low to begin with, but on the road to AGI, we should never be satisfied!😸
English
0
1
1
295
Hermes Agent Tips
Hermes Agent Tips@HermesAgentTips·
what makes Qwen 3.6 27B the best local LLM model compared to the others in its class?
English
43
0
83
14.2K
toriset
toriset@torisetxd·
@ZaiforStartups i think its actually the code quality, at the beginning the code is okay but it becomes sloppy very quickly once you stop trying to care because it takes a lot more work
English
0
0
3
885
Z.ai for Startups
Z.ai for Startups@ZaiforStartups·
The hardest problem in AI agents may no longer be intelligence. It’s coordination. Multi-agent systems are failing 41–87% of the time — mostly from coordination breakdowns, not model weakness. which means: the next infrastructure layer isn’t smarter models. It could be systems that keep agents aligned, verified, and on track.
English
32
18
417
36.4K
toriset
toriset@torisetxd·
@Elaina43114880 that post might be a teaser at it.. but its a pretty hard task while still maintaining the very niche knowledge gemini has, 50% is already a very good number imo
English
1
0
1
535
Logan Kilpatrick
Logan Kilpatrick@OfficialLoganK·
Why don’t LLM’s just tell you when you are asking a question / doing something that is out of distribution?
English
313
63
2K
228.2K
toriset retweetledi
Logan Kilpatrick
Logan Kilpatrick@OfficialLoganK·
The model is the product
English
209
71
1.9K
151.2K
toriset
toriset@torisetxd·
@bindureddy 50% cheaper would ruin their overall margins (not just gross/raw)
English
0
0
0
5
Bindu Reddy
Bindu Reddy@bindureddy·
The Gemini Pro model is rumored to be a GPT 5.5 level coding model 🧐 The catch - it will be more than 50% cheaper at $12/1M output token Gemini will take the lead over both GPT 5.5. and Opus 4.7 on the price-performance curve
English
141
33
1.1K
60.7K

Keşfet

@rattecs @stolsvik @teortaxesTex @trydotworks @xiong_hui_chen @alexworkmode @LukeParkerDev @sama