rvnx

37 ms·

931.
by rvnx 2y ago
Mixing politics and business is a terrible idea in that case. He had tons of money, lot of fun tech to play with, but wanted power and more women, now he became unofficial US president. So what will come next is people hating his decisions,
932.
by rvnx 2y ago
For now, I see (from the EU perspective) that the presidents of the US (Donald Trump and Elon Musk), are absolutely unreliable partners, and that the EU is alone with the problems. France population is never going to war, Germany also, Ital
933.
by rvnx 2y ago
The irony is that if we don't invest into defense, war would still likely break-out, but we may suffer less (or die less) because of a quick loss. Now we have a weak army in the EU, if Russia invades one of our friends, then the war wi
934.
Linux Running on an NES? [video](youtube.com)
1 points by rvnx 2y ago|0 comments
935.
by rvnx 2y ago
Love it, will check, thank you for showing / sharing all of that!
936.
by rvnx 2y ago
It is great discovery, it could even open a next step in AI with MoM "Mixture of Models", where small fine-tuned models take each part of a task (instead of the current MoE)
937.
by rvnx 2y ago
You sure ? Seeing Elon Musk, it could be a plausible move to screw financing for OpenAI. Without financing OpenAI may not have enough runway to survive. That would create a bigger market for x.AI, and also as a bonus act as a personal reven
938.
by rvnx 2y ago
You can choose as an investor between: a) Coca-Cola (~$300B, profitable, stable, stock-listed, well-known addictive brand, accepted business model with steady growth) b) OpenAI (bleeding cash: $5B lost last year, tech that is not really nov
939.
by rvnx 2y ago
Absolutely, we can even make a parallel with UNIX: UNIX was revolutionary. Companies poured billions into proprietary UNIX versions: IBM's AIX, Sun's Solaris, HP-UX, SGI IRIX... Then in 1991, Linus Torvalds released Linux for free
940.
by rvnx 2y ago
They care about the heavy censorship of the model, the amount of refusals and the low quality of the answers with the o1/o3 series.
941.
by rvnx 2y ago
This would have been true 2 years ago. Now with DeepSeek-R1, any company can host very similar service, so the technological edge has been lost. Google itself offers similar experience with Google Flash 2.0 Thinking. We haven't seen re
942.
by rvnx 2y ago
https://www.cnbc.com/2025/02/07/softbank-set-to-invest-40-bi... By sending the message that he wants to buy at 97B, he could imply that 260B is clearly exaggerated and create fear for these investors.
943.
by rvnx 2y ago
Most of all, this is free AI classification time
944.
by rvnx 2y ago
In the case they don't get high salaries from this activity, there is also a solution. The next step in ~10 years could be to offer their services to governments to offer "automated court decisions". Then the people who funde
945.
by rvnx 2y ago
I think you could reconsider DeepSeek-R1: it's actually really good. In comparison, o3-mini gets very vague in its reasoning, and gives surprisingly unhelpful answers (getting too short). Plus, let's not forget, R1 is available to
946.
by rvnx 2y ago
The last great one, before this abominable slow client that led to the exode from Skype.
947.
by rvnx 2y ago
https://x.com/zfescht/status/1886375804839313717 This is the best Tesla offers in Europe for now
948.
by rvnx 2y ago
Fast-forward in 10 years, we will have 128 GB of memory in our devices, better packed models and optimized LLM acceleration chips. Seems fine to run DeepSeek-R1 locally. So, no more OpenAI for day-to-day use unless a deep internet search is
949.
by rvnx 2y ago
Yes but only if you actually do the corrective actions, right ?
950.
by rvnx 2y ago
It's literally a browsing agent that searches the internet and they know the questions in advance when preparing the agent Without internet: 10% With internet: 23% In addition: > We found that the ground-truth answers for one datase
951.
by rvnx 2y ago
Doesn't make it untrue. Agents that can search the internet exist for a while now and have been essentially solved and happily used in platforms like Perplexity. It's really "meh", very far from revolutionary. Keep in mi
952.
by rvnx 2y ago
This is one of the actual questions: > In Greek mythology, who was Jason's maternal great-grandfather? https://www.google.com/search?q=In+Greek+mythology%2C+who+wa...
953.
by rvnx 2y ago
Especially this is not a breakthrough justifying a 340B USD valuation, but rather the work that junior developers can do; implement a loop of Bing Searches connected to an LLM.
954.
by rvnx 2y ago
In the past, they used another technique as well, they used corporate money to buy Dogecoin (and some Bitcoin too) to justify investments. Not in a small amount, we talk about more than 1B USD. https://techcrunch.com/2022&#x
955.
by rvnx 2y ago
44% of the people prefers the existing model ?
956.
by rvnx 2y ago
There is another advantage, you also avoid ads
957.
Tülu 3 405B – GPT-4o performance with open-source data, code, and recipes(twitter.com)
6 points by rvnx 2y ago|0 comments
958.
by rvnx 2y ago
As users, we would have more choices, more competition, and access to cheaper AI, more freedom (if DeepSeek stays open), higher quality models. The same with chips, if no sanctions on exports to NVIDIA, we would have access to Huawei Ascend
959.
by rvnx 2y ago
OpenAI already announced they will be copying DeepSeek implementation, and that they will make sure not to share it with anyone else than themselves
960.
by rvnx 2y ago
yes and as a bonus they may get sued, which in the long-term, makes free / offline models to not be viable It would be so much better if all models were trained with LibGen.

More ›