The 4 greatest AI tales from 2024 and one key prediction for 2025

Date:

Share post:

Be a part of our every day and weekly newsletters for the newest updates and unique content material on industry-leading AI protection. Study Extra


By all measures, 2024 was the largest 12 months for synthetic intelligence but — not less than on the subject of the commercialization of the know-how.

The massive language mannequin (LLM) growth spurred on by ChatGPT’s launch in late 2022 confirmed no indicators of slowing down, with quite a few new LLMs launched by not simply OpenAI and stalwart tech giants equivalent to Microsoft, Meta and Google, but additionally scores of different startups and particular person builders.

Stories of AI analysis slowing proved to be, if not unfounded, actually overstated for now.

As well as, new applied sciences past the Transformer structure that underpins most large LLMs started to floor, equivalent to Liquid AI’s Liquid Basis Fashions.

And eventually, corporations started to totally embrace the “agentic” method to AI — creating particular AI-powered bots, functions, and workflows that may work on particular issues independently, or with much less human stewardship than the everyday back-and-forth of LLM chatbots.

Distilling the 12 months’s information tales all the way down to a prime 14, a lot much less a prime 10 or prime 4, was a vexing effort. However I’ve gone forward and tried, albeit dishonest barely by combining a number of tales into bigger themes. In my eyes, right here’s what’s going to make the biggest impression shifting out from this 12 months:

1. OpenAI expanded far and broad past ChatGPT

The corporate arguably most chargeable for kicking off the gen AI period didn’t miss a beat this 12 months, regardless of intensifying competitors from newbies and legacy tech, even its personal investor and associate Microsoft.

o1 Mannequin: OpenAI launched its first new household of enormous general-purpose fashions past its GPT collection, the o1 “reasoning” collection, which allocates extra time to course of complicated prompts, leading to increased accuracy. It’s significantly efficient in science, coding and reasoning duties.

o3 Mannequin: It adopted the o1 mannequin from September with a blockbuster end-of-the-year announcement of a good extra superior o3 mannequin. Whereas this gained’t be availably publicly and even to any third events until early 2025, it reveals OpenAI shouldn’t be resting on its laurels.

ChatGPT Search: This characteristic, initially launched as an invitation-only standalone product referred to as SearchGPT earlier than being collapsed into ChatGPT correct, allows extra real-time internet data retrieval inside ChatGPT and a refined presentation of search outcomes, enhancing its utility for up-to-date queries and going head-to-head in opposition to Google, Bing and newcomer Perplexity.

Canvas: Launched in October, Canvas expands the ChatGPT interface past that of a conversational one to a workstation-like pane that may replace content material dynamically on the consumer’s request, equivalent to enhancing a doc or coding undertaking. In fact, it was arduous to not see it as a response to, or not less than a comparable characteristic to, Anthropic’s Artifacts introduced a number of months prior.

Sora: After practically a 12 months of teasing us with its intently guarded video generator mannequin, OpenAI in early December lastly launched Sora to the plenty, shortly inviting a variety of reactions because it sought to distinguish in a hotly aggressive AI video house with a singular and nicely thought-out interface and storyboarding characteristic.

2. Open-source AI took off

Llama 3 and three.1: Meta launched Llama 3 in April, setting a brand new normal for efficiency in open-source AI, then adopted it up shortly with Llama 3.1 in July with 405 billion parameters. Variations of Llama 3.1 had been used to energy Meta AI, the corporate’s assistant built-in throughout platforms like WhatsApp, Messenger, Instagram, and Fb, aiming to turn out to be probably the most broadly used AI assistant.

Llama 3.3: Launched in December 2024, Llama 3.3 delivered efficiency similar to bigger fashions however at a fraction of the computational value, making it extra accessible for enterprise functions.

In the meantime, Chinese language fashions equivalent to Alibaba’s Qwen-2.5 household and DeepSeek’s new V2.5 and R1-Lite Preview appeared seemingly out of nowhere to prime a number of the benchmark charts, and Nvidia itself went past supplying graphics playing cards and software program architectures to launch its personal open-source, highly effective Nemotron-70B mannequin.

Nous Analysis, a small outfit in San Francisco aiming to supply extra customized and fewer restrictive AI fashions as open-source, additionally debuted a number of cool new concepts.

And let’s not overlook France’s Mistral, which quickly expanded its personal open-source and proprietary AI choices.

3. Google’s Gemini collection turned a critical contender for the most effective obtainable

Within the comeback story of the 12 months, Google’s Gemini collection of AI fashions that had been as soon as mocked for his or her unusual picture generations and criticized for being overly “woke” got here again roaring with new, extra highly effective variations that now prime the third-party efficiency benchmark charts and are more and more interesting to builders and companies.

Google launched Gemini 2.0 Flash, a multimodal AI mannequin that helps streaming video evaluation and may see and instruct what you’re doing in your display screen, and adopted it up with Gemini 2.0 Flash Considering that competes with OpenAI’s o1 and o3 reasoning fashions.

4. Agentic AI swept the enterprise

Because the 12 months went on, “agentic” AI went from being a buzzworld to an actual collection of main product bulletins and initiatives by prime enteprise software program distributors. Take for instance:

Salesforce’s Agentforce 2.0: Salesforce unveiled Agentforce 2.0 a number of days in the past, a sophisticated AI agent program to boost reasoning, integration and customization options throughout its CRM and gross sales choices, in addition to Slack, considerably bettering enterprise productiveness instruments.

SAP’s Joule: SAP transformed its Joule chatbot into an AI agent powered by open-source massive language fashions (LLMs), driving innovation and effectivity in enterprise settings.

Google’s Challenge Astra: As a part of the Gemini 2.0 initiative, Google launched Challenge Astra, an AI assistant designed to supply real-time, contextual responses by leveraging Google’s suite of companies, aiming to boost consumer productiveness and decision-making.

My large prediction for 2025: AI-generated content material will reign supreme

Constructing upon these developments, 2025 is poised to witness the proliferation of AI-generated content material throughout enterprise and client domains, particularly as everybody from OpenAI to Meta, Google, Microsoft, Apple, and even Elon Musk’s xAI now has AI picture mills inbuilt to their choices.

This growth will streamline content material creation, improve personalization, and drive effectivity in numerous sectors.

Moreover, we anticipate the preliminary large-scale deployments of enormous language fashions (LLMs) and generative AI-powered robotics in each business and client settings, revolutionizing automation and human-robot interactions.

That’s all within the final #AIBeat publication for 2024. Thanks for studying, writing, subscribing, sharing, commenting, and for being right here with us. Trying ahead to sharing extra and listening to extra from you all in 2025.

Comfortable holidays and New Yr from all of us at VentureBeat to you and your family members.

Related articles

The code whisperer: How Anthropic’s Claude is altering the sport for software program builders

Be a part of our every day and weekly newsletters for the most recent updates and unique content...

Breakthrough T1D Play has raised $5M for diabetes analysis

The Breakthrough T1D Play program is a medical analysis charity elevating cash for important analysis into diabetes, one of many...

OpenAI’s o3 exhibits exceptional progress on ARC-AGI, sparking debate on AI reasoning

Be part of our each day and weekly newsletters for the most recent updates and unique content material...

Android telephone makers dropped the ball on Qi2 in 2024

Android telephones have been the primary to characteristic a bunch of notable requirements. They had been the primary...