Tag: llm

Gigabyte AI Top Unveiled at Computex 2024, to Enable End-to-End Local AI Training

Gigabyte unveiled its end-to-end artificial intelligence (AI) solution to train large language models (LLMs) locally on a device during AMD’s Computex 2024 event. The full-stack AI solution includes AI Top Utilities, AI Top Hardware, and AI Top Tutor, which encompass various aspects of training open-source AI models. AI Top Utilities is a training software with support for multiple open-source AI models, whereas AI Top Hardware offers the company’s AI-focused products. The AI Top Tutor is for those who require assistance in understanding how to make the most out of this solution.

Gigabyte AI Top offers solutions to train on-device AI

Making the announcement during the Computex 2024 event, the company unveiled AI Top as an all-encompassing solution that aims to “Train Your Own AI on Your Desk”. In a press release, the company detailed the various aspects of the offering which includes three divisions — software support, hardware support, and consultation and technical support. Notably, the announcement comes after the Gigabyte AI PC was introduced at Consumer Electronics Show (CES) 2024.

Gigabyte AI Top Utility
Photo Credit: Gigabyte

 

The AI Top Utility is a digital interface that allows local AI model training using new workflows. The company claims the software offers a user-friendly interface and real-time progress monitoring. It supports multiple open-source AI models with up to 236 billion parameters. The company claims the platform is more cost-effective and shows faster results compared to the cloud counterpart. It can also offload data to system memory and SSDs to surpass the limitations of VRAM size.

Next is the AI Top Hardware, which is essentially hardware offerings from the company. It features a series of AI optimised products that are power efficient and can handle AI training workloads. These hardware solutions also include upgradeable components. One of the primary hardware in this series includes AI Top Motherboard (TRX50) with configurable form factor, memory type and slots, graphics interface and more. AI Top Graphics Card, SSD, and PSU are also included.

The last offering in this solution is the AI Top Tutor. Positioning it as “on-desk AI coaching”, this is essentially the company’s AI-powered consultation and technical support system that offers insights, set-up guidance, and troubleshooting help. The company claims the coaching system will empower both beginners and professionals in starting on-device AI projects.

The company has not announced the pricing or availability of the Gigabyte AI Top solutions.

Check out our Latest News and Follow us at Facebook

Original Source

OpenAI Creates Safety and Security Committee as It Begins Testing Its Next Major AI Model

OpenAI announced the formation of a new committee to oversee the safety and security measures taken in the company’s projects and operations, on Tuesday. The Safety and Security Committee comprises select board members and is tasked with evaluating and further developing the San Francisco-based AI firm’s processes and safeguards. The new committee was formed as OpenAI has started testing the next generation of its artificial intelligence (AI) model. Notably, the company recently shared its Model Spec, which is a document that highlights the company’s approach towards building a responsible and ethical AI model.

OpenAI’s Safety and Security Committee

In a blog post, the company highlighted the details of its newly formed committee. The post stated, “Today, the OpenAI Board formed a Safety and Security Committee led by directors Bret Taylor (Chair), Adam D’Angelo, Nicole Seligman, and Sam Altman (CEO). This committee will be responsible for making recommendations to the full Board on critical safety and security decisions for OpenAI projects and operations.”

Apart from the directors, OpenAI’s Head of Preparedness Aleksander Madry, Head of Safety Systems John Schulman, Head of Security Matt Knight, and Chief Scientist Jakub Pachocki will also be a part of the committee.

The committee’s first task will be to evaluate and further develop the AI firm’s processes and safeguards over the next 90 days. After this, the committee will share its findings and recommendations with the full Board. The recommendations will go through the Board’s review, after which OpenAI will share the adopted recommendations publicly.

OpenAI starts testing its next generation AI model

These new developments at OpenAI are related to the company’s new initiative. It has started testing the next generation of its AI model. Calling it the “frontier” AI model, it highlighted that this under-testing large language model (LLM) will bring the company a step closer to Artificial General Intelligence (AGI).

AGI is a type of AI that can understand, learn, and apply knowledge across a wide range of real-world tasks comparable to human intelligence. Some definitions of AGI also highlight that it is capable of autonomous functioning and can develop some level of self-awareness.

Affiliate links may be automatically generated – see our ethics statement for details.

For the latest tech news and reviews, follow Gadgets 360 on X, Facebook, WhatsApp, Threads and Google News. For the latest videos on gadgets and tech, subscribe to our YouTube channel. If you want to know everything about top influencers, follow our in-house Who’sThat360 on Instagram and YouTube.


Apple, Google, Amazon Lobby Group Opposes India’s EU-Like Antitrust Proposal



Check out our Latest News and Follow us at Facebook

Original Source

Alibaba, Baidu Slash Prices of Large-Language Models Used to Power AI Chatbots

Chinese tech giants Alibaba and Baidu slashed prices on Tuesday of large-language models (LLMs) used to power generative artificial intelligence products, as a price war in the cloud computing sector heats up in China.

Alibaba’s cloud unit announced price cuts of up to 97 percent on a range of its Tongyi Qwen LLMs. Its Qwen-Long model, for instance, will cost only CNY 0.0005 per 1,000 tokens – or units of data processed by the LLM – after the price cut, down from CNY 0.02 per 1,000 tokens.

It was quickly followed by Baidu, which hours later announced that its Ernie Speed and Ernie Lite models would be free for all business users.

A price war in China’s cloud computing space has been ongoing for the past few months, with Alibaba and Tencent recently lowering prices of their cloud computing services.

Many Chinese cloud vendors have relied on AI chatbot services to boost sales, after China saw a wave of investment in large language models in response to the hit debut of US-based OpenAI’s ChatGPT in late 2022.

The price war in China’s cloud computing space has now hit the large-language models that power these chatbots, threatening to lower companies’ profit margins.

Baidu’s Ernie Lite and Ernie Speed were released in March and until Tuesday corporate customers paid to use them.

Bytedance announced last week that the main model of its Doubao LLMs would be priced 99.3% lower than the industry average for business users.

Chinese LLM developers have focused on charging businesses as a way to monetize their investments in LLMs.

Some have also begun targeting individual users. Chinese startup Moonshot recently launched a tipping feature, where business and individual users can pay to prioritize their use of its chatbot services.

Baidu was the first company in China to offer its LLM products to paying consumers, charging 59 yuan per month for those looking to use its most advanced Ernie 4 model.

© Thomson Reuters 2024

Affiliate links may be automatically generated – see our ethics statement for details.

Check out our Latest News and Follow us at Facebook

Original Source

OpenAI GPT-4o Begins Rolling Out to Some Users, Gets Web Searching Capability

OpenAI GPT-4o artificial intelligence (AI) model was unveiled on Monday, and it is now being rolled out to some users. The newest flagship-grade AI model by the company introduced significant improvements in the speech and vision capabilities of the chatbot, as well as added a better understanding of the language and context of the queries. For now, users are getting the AI model with limited access and the voice and video features are not available to use. People can, however, use its text and web search capabilities.

Gadgets 360 got access to the GPT-4o model on Friday morning. This confirms that the AI model will be available in India, even when it was not specified during OpenAI’s Spring Update event. However, it was available to only a couple of staff members, so it’s likely the company is rolling out the AI model gradually and it will take a few weeks before everyone can use it. The limited access is also very restrictive. We were able to get about ten questions in before our limit expired, and then we were shifted back to GPT-3.5.

ChatGPT’s GPT-4o – website view

Currently, users cannot do anything to get access faster than others. There is no waitlist to join. However, users will require an OpenAI account to be eligible for this update. Once GPT-4o is available, users will get a message when opening the website which mentions that they can now access it in a limited capacity. If you have the same account on your Android or iOS app, you will get access to the model there as well. Do note, that reloading the page will make the message disappear.

There is an easy test to check whether you have GPT-4o or not (in case you might have missed the message). After opening the ChatGPT website, you can see a collapsible menu on the top left within the margin. If you do not have access to GPT-4o, and you’re a free user, it will show ChatGPT 3.5 and give you the option to sign up for ChatGPT Plus with access to GPT-4. However, if you do have access to the new AI model, the menu will not mention any numbers and simply mention ChatGPT and ChatGPT Plus. Further, the lightning icon is replaced with two intersecting elliptical circles (looks like a minimalist atom icon).

ChatGPT’s GPT-4o – Android app view

We took the new AI model for a spin and found some improvements in its responses. One particular use case was in solving mathematical equations. Compared to GPT-3.5, it now shows the answers in a better format and does not complete multiple steps in one go. Creative generation is also more fluid and the usage of its ‘robotic’ language has reduced significantly. And the biggest upgrade is that this model can search the web to give you the latest information, so you do not have to worry about its knowledge cut-off anymore. Every web-based search result now comes with citations for which website was used to get the information.

Check out our Latest News and Follow us at Facebook

Original Source

Mistral Launches 8X22B Mixture of Experts AI Model in Open Source, Benchmarks Reveal Big Upgrades

Mistral released its latest artificial intelligence (AI) model, 8X22B, on Wednesday. The mixture-of-expert AI model follows the 8X7B model released in December 2023 and offers a larger parameter size. The company known for its fully open-source AI models released the latest model unconventionally, without any announcement posts or blog posts accompanying it. While there are no benchmarks provided by the AI firm itself, the Hugging Face community users tested it and posted benchmark scores of the 8X22B model. The results appear to close the gap with the closed models from OpenAI and Google.

The official X (formerly known as Twitter) account of Mistral released the 8X22B AI model via a torrent magnet link, continuing its unconventional method of dropping AI models without an announcement and directly for people to download. Mistral is also one of the only truly open-source platforms that not only makes the weights open but also the entire architecture. However, it should be noted that most devices are not equipped to run powerful AI models on-device, and it might cause the device to malfunction. The total file size is 262GB.

Mistral’s 8X22B is an autocomplete AI model. These are typically different from instruct or chat variants of AI models. OpenAI’s ChatGPT and Google’s Gemini AI are chat models, whereas Meta’s Code Llama 7B and 13B are instruct models. Chat models are those which can understand natural language and contextual queries to provide the correct response. Instruct models are mainly used by developers who ask the AI model to perform a specific task. In contrast, an autocomplete model completes the sentence that has been provided in the prompt.

Early benchmark scores based on a Hugging Face community post show that the latest Mistral model offers significant upgrades over its predecessors and closes the gap with other major models. In the Hellaswag benchmark, Mistral 8X22B scored 88.9, closely trailing GPT-4 at 95.3, Claude 3 Opus at 95.4, and Gemini 1.5 Pro at 92.5. However, it outscores GPT-3.5 at 85.5 and Gemini 1.0 Ultra at 87.8.

Comparing the benchmarks with Mistral’s own models, the 8X22B falls between Mistral-medium and Mistral-large models. With a later release of an instruct or chat variant, it might even fare similarly to Mistral-large.

Affiliate links may be automatically generated – see our ethics statement for details.

Comments

For the latest tech news and reviews, follow Gadgets 360 on X, Facebook, WhatsApp, Threads and Google News. For the latest videos on gadgets and tech, subscribe to our YouTube channel. If you want to know everything about top influencers, follow our in-house Who’sThat360 on Instagram and YouTube.


Apple Warns iPhone Users of ‘Mercenary Spyware’ Attack in 92 Countries


Motorola Edge 50 Ultra Appears on Geekbench, May Run on Snapdragon 8s Gen 3 SoC



Check out our Latest News and Follow us at Facebook

Original Source

Fireworks.ai, the Generative AI Firm That Fine-Tunes and Customises Open-Source LLMs For Business Needs

Fireworks.ai is a California-based artificial intelligence (AI) startup that is offering a unique solution for enterprises. The AI firm does not build large language models (LLMs) or foundation models from scratch but fine-tunes open-source models and converts them into an Application Programming Interface (API) to help businesses deploy the AI capabilities in a seamless fashion. The fine-tuning reduces the scope of the AI model and focuses it on a specific functionality. This allows them to reduce instances of AI hallucinations and improve the capabilities of the model significantly.

The AI firm was co-founded by Lin Qiao who also holds the seat of the CEO in the company. After serving as the Senior Director of Engineering at Meta and working with AI frameworks and platforms, Qiao and her team founded the startup in October 2022, as per her LinkedIn profile. In a conversation with TechCrunch, she explained the business model of Fireworks.ai, highlighting the fine-tuning service they provide. She said, “It can be either off the shelf, open source models or the models we tune or the models our customer can tune by themselves. All three varieties can be served through our inference engine API.”

This puts the firm in a unique position where while it is not innovating at the foundation model level, it is bridging the gap between an LLM and a business-ready product that can be deployed seamlessly. With a primary focus on building APIs, Fireworks.ai lets its enterprise clients plug and play any open-source AI model in its catalogue. As per the report, the company also lets businesses experiment with different AI models to choose the one that fits their needs.

At present, the startup claims to contain 89 open-source LLMs such as Mixtral MoE 8x7B Instruct, Meta’s Llama 2 70B Chat, Google’s Gemma 7B Instruct, Stability AI’s Stable Diffusion XL, and more. The AI firm offers the models in either serverless format that does not require businesses to configure hardware or deploy models, or as on-demand models which are available for dedicated deployments, served on reserved GPU configurations according to business needs.

For the on-demand format, Fireworks.ai has three payment plans — Developer, Business, and Enterprise — where the Developer plan comes with a pay-per-usage structure and a rate limit of 600 requests per minute, the Enterprise tier has custom pricing offers and unlimited rate limits. The serverless format is billed at a per-token pricing plan where different models, depending on whether they are text-only, image-only, or multimodal, will fetch a different price.

Affiliate links may be automatically generated – see our ethics statement for details.

Check out our Latest News and Follow us at Facebook

Original Source

Baidu Launches $145 Million Venture Capital Fund for Startups Working on AI

Chinese search giant Baidu will set up a venture capital fund of CNY 1 billion (roughly Rs. 1,200 crore) to back start-ups focused on content generated by artificial intelligence applications, it said on Wednesday.

The company will also launch a competition for developers to build applications off its ERNIE large language model (LLM) or integrate the model into their existing products, it added.

Chinese tech companies have raced to release their own LLMs following the dramatic success of ChatGPT, the AI-powered chatbot released by Microsft-backed OpenAI.

Almost 80 organisations in China have launched their own LLMs since 2020, with releases this year slightly exceeding those of the United States, a report showed this week.

In March, Baidu unveiled Ernie Bot, its own AI-powered LLM. E-commerce giant Alibaba Group Holding was among the other Chinese companies that followed quickly.

In response to the surge of LLMs, China published draft regulations in April on the use of generative AI.

© Thomson Reuters 2023

Samsung Galaxy A34 5G was recently launched by the company in India alongside the more expensive Galaxy A54 5G smartphone. How does this phone fare against the Nothing Phone 1 and the iQoo Neo 7? We discuss this and more on Orbital, the Gadgets 360 podcast. Orbital is available on Spotify, Gaana, JioSaavn, Google Podcasts, Apple Podcasts, Amazon Music and wherever you get your podcasts.
Affiliate links may be automatically generated – see our ethics statement for details.

Check out our Latest News and Follow us at Facebook

Original Source

Exit mobile version