This week in artificial intelligence: Perhaps we must ignore artificial intelligence standards at the present time

Welcome to the regular newsletter of Techcrunch! We go to a large extent, but you can find all our artificial intelligence coverage, including columns, our daily analysis, urgent news stories, in Techcrunch. If you want these stories and much more in your inbox every day, subscribe to our daily news messages here.

This week, Elon Musk’s Ai Startup, Xai, the latest model of pioneering artificial intelligence, released, Groc 3That operate the company’s GROK Chatbot applications. The model has been trained on about 200,000 graphics processing units, outperforming a number of other leading models, including from Openai, on mathematics, programming and more standards.

But what really tell us these criteria?

Here in TC, we often reluctantly report the records because it is one of the relatively few unified methods (relatively) that the artificial intelligence industry measures model improvements. The famous artificial intelligence standards tend to test Ethical knowledge, giving total degrees that are badly related to efficiency In the tasks that most people are interested in.

Professor Ethan Malik in Warton also indicated A series of posts on X After unveiling GROK 3 on Monday, there is “an urgent need for better batteries than independent tests and tests.” AI Companes Select Sendark is often related to Mollick, making these results more striking to accept them with the nominal value.

“The general standards are both” Mah “and are saturated, which leaves a lot of artificial intelligence tests to be like food reviews, based on taste.” “If artificial intelligence is very important to work, we need more.”

There is no shortage of independent Tests and Organizations Suggesting new standards for Amnesty International, but its relative merit is far from the stable issue in the industry. Some commentators and experts suggest artificial intelligence Standards of standards with economic impact To ensure its benefit, while Others argue that adoption and interest They are the final standards.

This discussion may be angry until the end of time. Maybe we must instead, It also describes the X Roon userYou have to pay less attention to new models and standards that prohibit the main artistic breakthroughs of Amnesty International. For our collective mind, this may not be the worst idea, even if it stimulates a level of Ai Fomo.

As mentioned above, this week in artificial intelligence occurs in a stop. Thanks for committing to us, readers, through this rolling ship on a trip. Until the next time.

news

Openai tries to “Uncensor” Chatgpt: Max wrote about how to change Openai to the approach to developing artificial intelligence to adopt “intellectual freedom”, regardless of the extent of challenge or controversy.

The start of the new Mira: Starting the former new Openai CTO Mirati, Thinking machines laboratoryIt intends to build tools “to make Amnesty International work to meet the unique needs and goals (for people).”

GROK 3 Comment: Elon Musk’s Ai Startup, Xai, the latest pioneering AI, GROK 3, has released new possibilities for iOS and web applications.

Lama conference: Meta will host the first developers developed for the tweeted artificial intelligence relationship this spring. Llamacon is called after the Llama family in Meta, from the Truidic IQ, and the conference is scheduled to be held on April 29.

Amnesty International and digital sovereignty in Europe: Paul has appointed Openeurlm, a cooperation between about 20 organizations to build a “series of basic models for AI transparent in Europe” that maintains “linguistic and cultural diversity” for all the languages ​​of the European Union.

Search paper in the week

Openai researchers have created a new standard for Amnesty International, Swe-LancerThis aims to assess the ingenuity of the coding of strong artificial intelligence systems. The standard consists of more than 1,400 independent software engineering tasks ranging from error repairs and spreading features to technical implementation proposals “at the level of the manager”.

According to Openai, the best performance artificial intelligence model, Claude 3.5 Sonnet of the human being, records 40.3 % on the full Swe-Lancer-which indicates that artificial intelligence has great ways to go. It should be noted that researchers did not evaluate modern models such as Openai’s O3-Mini Or the Chinese company AI Deepseek’s R1.

Week model

A Chinese company of artificial intelligence, called Stepfun, released an “open” model for Amnesty International. stepHe can understand and generate speech in several languages. It supports STEP-UADIO, English, and Japanese and allows users to control emotion and even the artificial sound tone it creates, including singing.

STEPFUN is one of many well -funded Chinese startup companies to launch models under a lenient license. Founded in 2023, Stepfun According to what was reported recently A financing round worth several hundred million dollars from a group of investors that include the state -owned Chinese private stock companies.

Grab a bag

Nous Research, an Amnesty International Research Group, Absolute What you claim is one of the first artificial intelligence models that explain logic and “the possibilities of the intuitive language model”.

The model can preview Deephermes-3, switch and stop long “intellectual chains” to improve accuracy at the expense of some arithmetic weight. In “Thinking” mode, Deephermes-3, similar to other artificial intelligence models, “believed” for a longer period of the most difficult problems and shows that its thinking process reaches the answer.

It was reported that Antarbur You plan to issue an architectural similar model soonOpenai said this model On the near -term road map.