In partnership with Tsinghua University researchers, DeepSeek has created a technique detailed in a research paper, titled “Inference-Time Scaling for Generalist Reward Modeling.” It outlines how a new approach outperforms existing methods and how the team “achieved competitive performance” compared to strong public reward models.

The innovation focuses on enhancing how AI systems learn from human preferences – a important aspect of creating more useful and aligned artificial intelligence.

What are AI reward models, and why do they matter?

AI reward models are important components in reinforcement learning for large language models. They provide feedback signals that help guide an AI’s behaviour toward preferred outcomes. In simpler terms, reward models are like digital teachers that help AI understand what humans want from their responses.

“Reward modeling is a process that guides an LLM towards human preferences,” the DeepSeek paper states. Reward modeling becomes important as AI systems get more sophisticated and are deployed in scenarios beyond simple question-answering tasks.

The innovation from DeepSeek addresses the challenge of obtaining accurate reward signals for LLMs in different domains. While current reward models work well for verifiable questions or artificial rules, they struggle in general domains where criteria are more diverse and complex.

The dual approach: How DeepSeek’s method works

DeepSeek’s approach combines two methods:

1. Generative reward modeling (GRM): This approach enables flexibility in different input types and allows for scaling during inference time. Unlike previous scalar or semi-scalar approaches, GRM provides a richer representation of rewards through language.
2. Self-principled critique tuning (SPCT): A learning method that fosters scalable reward-generation behaviours in GRMs through online reinforcement learning, one that generates principles adaptively.

One of the paper’s authors from Tsinghua University and DeepSeek-AI, Zijun Liu, explained that the combination of methods allows “principles to be generated based on the input query and responses, adaptively aligning reward generation process.”

The approach is particularly valuable for its potential for “inference-time scaling” – improving performance by increasing computational resources during inference rather than just during training.

The researchers found that their methods could achieve better results with increased sampling, letting models generate better rewards with more computing.

Implications for the AI Industry

DeepSeek’s innovation comes at an important time in AI development. The paper states “reinforcement learning (RL) has been widely adopted in post-training for large language models […] at scale,” leading to “remarkable improvements in human value alignment, long-term reasoning, and environment adaptation for LLMs.”

The new approach to reward modelling could have several implications:

1. More accurate AI feedback: By creating better reward models, AI systems can receive more precise feedback about their outputs, leading to improved responses over time.
2. Increased adaptability: The ability to scale model performance during inference means AI systems can adapt to different computational constraints and requirements.
3. Broader application: Systems can perform better in a broader range of tasks by improving reward modelling for general domains.
4. More efficient resource use: The research shows that inference-time scaling with DeepSeek’s method could outperform model size scaling in training time, potentially allowing smaller models to perform comparably to larger ones with appropriate inference-time resources.

DeepSeek’s growing influence

The latest development adds to DeepSeek’s rising profile in global AI. Founded in 2023 by entrepreneur Liang Wenfeng, the Hangzhou-based company has made waves with its V3 foundation and R1 reasoning models.

The company upgraded its V3 model (DeepSeek-V3-0324) recently, which the company said offered “enhanced reasoning capabilities, optimised front-end web development and upgraded Chinese writing proficiency.” DeepSeek has committed to open-source AI, releasing five code repositories in February that allow developers to review and contribute to development.

While speculation continues about the potential release of DeepSeek-R2 (the successor to R1) – Reuters has speculated on possible release dates – DeepSeek has not commented in its official channels.

What’s next for AI reward models?

According to the researchers, DeepSeek intends to make the GRM models open-source, although no specific timeline has been provided. Open-sourcing will accelerate progress in the field by allowing broader experimentation with reward models.

As reinforcement learning continues to play an important role in AI development, advances in reward modelling like those in DeepSeek and Tsinghua University’s work will likely have an impact on the abilities and behaviour of AI systems.

Work on AI reward models demonstrates that innovations in how and when models learn can be as important increasing their size. By focusing on feedback quality and scalability, DeepSeek addresses one of the fundamental challenges to creating AI that understands and aligns with human preferences better.

See also: DeepSeek disruption: Chinese AI innovation narrows global technology divide

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek’s AIs: What humans really want appeared first on AI News.

The new model advanced seven points in the benchmark to surpass proprietary counterparts such as Google’s Gemini 2.0 Pro, Anthropic’s Claude 3.7 Sonnet, and Meta’s Llama 3.3 70B.

While V3-0324 trails behind reasoning models, including DeepSeek’s own R1 and offerings from OpenAI and Alibaba, the achievement highlights the growing viability of open-source solutions in latency-sensitive applications where immediate responses are critical.

DeepSeek V3-0324 represents a new era for open-source AI

Non-reasoning models – which generate answers instantly without deliberative “thinking” phases – are essential for real-time use cases like chatbots, customer service automation, and live translation. DeepSeek’s latest iteration now sets the standard for these applications, eclipsing even leading proprietary tools.

“This is the first time an open weights model is the leading non-reasoning model, a milestone for open source,” states Artificial Analysis. The model’s performance edges it closer to proprietary reasoning models, though the latter remain superior for tasks requiring complex problem-solving.

DeepSeek V3-0324 retains most specifications from its December 2024 predecessor, including:  

• 128k context window (capped at 64k via DeepSeek’s API)
• 671 billion total parameters, necessitating over 700GB of GPU memory for FP8 precision
• 37 billion active parameters
• Text-only functionality (no multimodal support) 
• MIT License

“Still not something you can run at home!” Artificial Analysis quips, emphasising its enterprise-grade infrastructure requirements.

Open-source AI is bringing the heat

While proprietary reasoning models like DeepSeek R1 maintain dominance in the broader Intelligence Index, the gap is narrowing.

Three months ago, DeepSeek V3 nearly matched Anthropic’s and Google’s proprietary models but fell short of surpassing them. Today, the updated V3-0324 not only leads open-source alternatives but also outperforms all proprietary non-reasoning rivals.

“This release is arguably even more impressive than R1,” says Artificial Analysis.

DeepSeek’s progress signals a shift in the AI sector, where open-source frameworks increasingly compete with closed systems. For developers and enterprises, the MIT-licensed V3-0324 offers a powerful, adaptable tool—though its computational costs may limit accessibility.

“DeepSeek are now driving the frontier of non-reasoning open weights models,” declares Artificial Analysis.

With R2 on the horizon, the community awaits another potential leap in AI performance.

(Photo by Paul Hanaoka)

See also: Hugging Face calls for open-source focus in the AI Action Plan

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek V3-0324 tops non-reasoning AI models in open-source first appeared first on AI News.

I wonder whether that may be partly a generational thing. The baby boomer generation was the first to be widely employed in IT and that cohort learned the lessons of business the hard way.  Projects had to be cost-justified because technology was expensive and needed to be attached to a robust ROI case. Projects were rolled out slowly because they were complex and had to be aligned to a specific business need, endorsed by the right stakeholders. ‘Project creep’ was feared and the relationship between IT and ‘the business’ was often fraught and complex, characterised by mutual suspicion.

Today, the situation is somewhat different. The IT industry is enormous, the Fortune 50 is replete with major tech brands and other sectors marvel at the profit margins of the software sector. That may all be very well for Silicon Valley and the venture capitalists of Sand Hill Road desperate to find The Next Big Thing. But back in the real world of corporate IT, matters should be seen with more caution, an appropriate level of pragmatism and even a raised eyebrow or two.

Which brings us back to AI. AI is far from new and has its roots all the way back in the middle of the previous century. So far, despite all the excitement, it has played only a moderate role in the business world. The success of tools like Chat-GPT has catapulted it to mainstream attention but it is still beset by familiar issues. It is costly to deploy in earnest, it requires (at least until DeepSeek) enormous compute power to develop and it delivers responses that are often questionable. There are also serious questions to be asked about legal liability and copyright.

A balancing act

We need to strike a happy balance between the boosterism and experimentation inherent in AI today and a healthy sense of pragmatism. We should begin with the business case and ask how AI helps us. What is our mission? Where are our strategic opportunities and risks? OK, now how can AI help us? Today, there is too much “AI is great, let’s see what we can do with it”.

Today, I see AI as a massive opportunity but use cases need to be worked out. AI is great at massive computation tasks that human beings are bad at. It can study patterns and detect trends faster than our feeble human brains can. It doesn’t get out of the bed on the wrong side in the morning, tire easily or require two weeks holiday in the Mediterranean each year. It is surprisingly excellent at a limited number of creative tasks such as making images, music, poems and videos. But it is bad at seeing the big picture. It lacks the human sense of caution that keeps us from danger, and it has no experience of the real world of work that is composed of an enormous range of variables, not the least of which is human mood and perception.

AI today is great at the edge: in powering bots that answer predictable questions or agents that help us achieve rote tasks faster than would otherwise be the case. Robotic process automation has been a useful aid and has changed the dynamic of how the human being interacts with computers: we can now hand off dull jobs like processing credit card applications or expense claims and focus on being creative thinkers.

There are grey areas too. Conversational AI is a work in progress, but we can expect rapid improvements based on iterative continuous learning by our binary friends. Soon we may be impressed by AI’s ability to guess our next steps and to suggest smarter ways to accomplish our work. Similarly, there is scope for AI to learn more about our vertical businesses and to understand trends that humans may miss when we fail to see the forest for the trees.

But we are some way off robot CEOs, and we need to ensure that AI ‘decisions’ are tempered by human bosses that have common sense, the ability to check, test and revert. The future is one where AI and humanity work in concert but for now we are wise to deploy with care and with sensible budgets and the appropriate level of commitment.

We need to watch carefully for the next DeepSeek hit, query it and always begin with old-fashioned questions as to applicability, costs and risk. I note that DeepSeek’s website bears the tagline “Into the Unknown”. That’s about right: we need to maintain a spirit of adventure and optimism but avoid getting lost in a new technological wilderness.

• For more information, visit https://www.unit4.com/, follow Unit4 on Facebook: Unit4 Business Software, Instagram: @unit4global or visit its YouTube channel: Unit4 and LinkedIn page.

Photo by Solen Feyissa on Unsplash

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek is a reminder to approach the AI unknown with caution appeared first on AI News.

However, this development comes against the backdrop of mounting controversies that have drawn parallels to the TikTok saga.

Today, DeepSeek shared its intentions in a tweet that outlined its vision of open collaboration: “We’re a tiny team at DeepSeek exploring AGI. Starting next week, we’ll be open-sourcing five repos, sharing our small but sincere progress with full transparency.”

The repositories – which the company describes as “documented, deployed, and battle-tested in production” – include fundamental building blocks of DeepSeek’s online service.

By open-sourcing its tools, DeepSeek hopes to contribute to the broader AI research community.

“As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey. No ivory towers – just pure garage-energy and community-driven innovation,” the company said.

This philosophy has drawn praise for fostering collaboration in a field that often suffers from secrecy, but DeepSeek’s rapid rise has also raised eyebrows.

Despite being a small team with a mission rooted in transparency, the company has been under intense scrutiny amid allegations of data misuse and geopolitical entanglements.

Rising fast, under fire

Practically unknown until recently, DeepSeek burst onto the scene with a business model that stood in stark contrast to more established players like OpenAI and Google.

Offering its advanced AI capabilities for free, DeepSeek quickly gained global acclaim for its cutting-edge performance. However, its exponential rise has also sparked debates about the trade-offs between innovation and privacy.

US lawmakers are now pushing for a ban on DeepSeek after security researchers found the app transferring user data to a banned state-owned company.

A probe has also been launched by Microsoft and OpenAI over a breach of the latter’s systems by a group allegedly linked to DeepSeek.

Concerns about data collection and potential misuse have triggered comparisons to the controversies surrounding TikTok, another Chinese tech success story grappling with regulatory pushback in the West.

DeepSeek continues AGI innovation amid controversy

DeepSeek’s commitment to open-source its technology appears timed to deflect criticism and reassure sceptics about its intentions.

Open-sourcing has long been heralded as a way to democratise technology and increase transparency, and DeepSeek’s “daily unlocks,” that are set to begin soon, could offer the community reassuring insight into its operations.

Nevertheless, questions remain over how much of the technology will be open for scrutiny and whether the move is an attempt to shift the narrative amid growing political and regulatory pressure.

It’s unclear whether this balancing act will be enough to satisfy lawmakers or deter critics, but one thing is certain: DeepSeek’s open-source leap marks another turn in its dramatic rise.

While the company’s motto of “garage-energy and community-driven innovation” resonates with developers eager for open collaboration, its future may rest as much on its ability to address security concerns as on its technical prowess.

(Photo by Solen Feyissa)

See also: DeepSeek’s AI dominance expands from EVs to e-scooters in China

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including IoT Tech Expo, Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek to open-source AGI research amid privacy concerns appeared first on AI News.

DeepSeek, practically unknown just weeks ago, took the tech world by storm—gaining global acclaim for its cutting-edge performance while sparking debates reminiscent of the TikTok saga.

Its rise has been fuelled in part by its business model: unlike many of its American counterparts, including OpenAI and Google, DeepSeek offered its advanced powers for free.

However, concerns have been raised about DeepSeek’s extensive data collection practices and a probe has been launched by Microsoft and OpenAI over a breach of the latter’s system by a group allegedly linked to the Chinese AI startup.

A threat to US AI dominance

DeepSeek’s astonishing capabilities have, within a matter of weeks, positioned it as a major competitor to American AI stalwarts like OpenAI’s ChatGPT and Google Gemini. But, alongside the app’s prowess, concerns have emerged over alleged ties to the Chinese Communist Party (CCP).  

According to security researchers, hidden code within DeepSeek’s AI has been found transmitting user data to China Mobile—a state-owned telecoms company banned in the US. DeepSeek’s own privacy policy permits the collection of data such as IP addresses, device information, and, most alarmingly, even keystroke patterns.

Such findings have led to bipartisan efforts in the US Congress to curtail DeepSeek’s influence, with lawmakers scrambling to protect sensitive data from potential CCP oversight.

Reps. Darin LaHood (R-IL) and Josh Gottheimer (D-NJ) are spearheading efforts to introduce legislation that would prohibit DeepSeek from being installed on all government-issued devices. 

Several federal agencies, among them NASA and the US Navy, have already preemptively issued a ban on DeepSeek. Similarly, the state of Texas has also introduced restrictions.

Potential ban of DeepSeek a TikTok redux?

The controversy surrounding DeepSeek bears similarities to debates over TikTok, the social video app owned by Chinese company ByteDance. TikTok remains under fire over accusations that user data is accessible to the CCP, though definitive proof has yet to materialise.

In contrast, DeepSeek’s case involves clear evidence, as revealed by cybersecurity investigators who identified the app’s unauthorised data transmissions. While some might say DeepSeek echoes the TikTok controversy, security experts argue that it represents a much starker and documented threat.

Lawmakers around the world are taking note. In addition to the US proposals, DeepSeek has already faced bans from government systems in countries including Australia, South Korea, and Italy.  

AI becomes a geopolitical battleground

The concerns over DeepSeek exemplify how AI has now become a geopolitical flashpoint between global superpowers—especially between the US and China.

American AI firms like OpenAI have enjoyed a dominant position in recent years, but Chinese companies have poured resources into catching up and, in some cases, surpassing their US competitors.  

DeepSeek’s lightning-quick growth has unsettled that balance, not only because of its AI models but also due to its pricing strategy, which undercuts competitors by offering the app free of charge. That begs the question of whether it’s truly “free” or if the cost is paid in lost privacy and security.

China Mobile’s involvement raises further eyebrows, given the state-owned telecom company’s prior sanctions and prohibition from the US market. Critics worry that data collected through platforms like DeepSeek could fill gaps in Chinese surveillance activities or even potential economic manipulations.

A nationwide DeepSeek ban is on the cards

If the proposed US legislation is passed, it could represent the first step toward nationwide restrictions or an outright ban on DeepSeek. Geopolitical tension between China and the West continues to shape policies in advanced technologies, and AI appears to be the latest arena for this ongoing chess match.  

In the meantime, calls to regulate applications like DeepSeek are likely to grow louder. Conversations about data privacy, national security, and ethical boundaries in AI development are becoming ever more urgent as individuals and organisations across the globe navigate the promises and pitfalls of next-generation tools.  

DeepSeek’s rise may have, indeed, rattled the AI hierarchy, but whether it can maintain its momentum in the face of increasing global pushback remains to be seen.

(Photo by Solen Feyissa)

See also: AVAXAI brings DeepSeek to Web3 with decentralised AI agents

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek ban? China data transfer boosts security concerns appeared first on AI News.

According to Bloomberg, the investigation stems from suspicious data extraction activity detected in late 2024 via OpenAI’s application programming interface (API), sparking broader concerns over international AI competition.

Microsoft, OpenAI’s largest financial backer, first identified the large-scale data extraction and informed the ChatGPT maker of the incident. Sources believe the activity may have violated OpenAI’s terms of service, or that the group may have exploited loopholes to bypass restrictions limiting how much data they could collect.

DeepSeek has quickly risen to prominence in the competitive AI landscape, particularly with the release of its latest model, R-1, on 20 January.

Billed as a rival to OpenAI’s ChatGPT in performance but developed at a significantly lower cost, R-1 has shaken up the tech industry. Its release triggered a sharp decline in tech and AI stocks that wiped billions from US markets in a single week.

David Sacks, the White House’s newly appointed “crypto and AI czar,” alleged that DeepSeek may have employed questionable methods to achieve its AI’s capabilities. In an interview with Fox News, Sacks noted evidence suggesting that DeepSeek had used “distillation” to train its AI models using outputs from OpenAI’s systems.

“There’s substantial evidence that what DeepSeek did here is they distilled knowledge out of OpenAI’s models, and I don’t think OpenAI is very happy about this,” Sacks told the network.  

Model distillation involves training one AI system using data generated by another, potentially allowing a competitor to develop similar functionality. This method, when applied without proper authorisation, has stirred ethical and intellectual property debates as the global race for AI supremacy heats up.  

OpenAI declined to comment specifically on the accusations against DeepSeek but acknowledged the broader risk posed by model distillation, particularly by Chinese companies.  

“We know PRC-based companies — and others — are constantly trying to distill the models of leading US AI companies,” a spokesperson for OpenAI told Bloomberg.  

Geopolitical and security concerns  

Growing tensions around AI innovation now extend into national security. CNBC reported that the US Navy has banned its personnel from using DeepSeek’s products, citing fears that the Chinese government could exploit the platform to access sensitive information.

In an email dated 24 January, the Navy warned its staff against using DeepSeek AI “in any capacity” due to “potential security and ethical concerns associated with the model’s origin and usage.”

Critics have highlighted DeepSeek’s privacy policy, which permits the collection of data such as IP addresses, device information, and even keystroke patterns—a scope of data collection considered excessive by some experts.

Earlier this week, DeepSeek stated it was facing “large-scale malicious attacks” against its systems. A banner on its website informed users of a temporary sign-up restriction.

The growing competition between the US and China in particular in the AI sector has underscored wider concerns regarding technological ownership, ethical governance, and national security.  

Experts warn that as AI systems advance and become increasingly integral to global economic and strategic planning, disputes over data usage and intellectual property are only likely to intensify. Accusations such as those against DeepSeek amplify alarm over China’s rapid development in the field and its potential quest to bypass US-led safeguards through reverse engineering and other means.  

While OpenAI and Microsoft continue their investigation into the alleged misuse of OpenAI’s platform, businesses and governments alike are paying close attention. The case could set a precedent for how AI developers police model usage and enforce terms of service.

For now, the response from both US and Chinese stakeholders highlights how AI innovation has become not just a race for technological dominance, but a fraught geopolitical contest that is shaping 21st-century power dynamics.

(Image by Mohamed Hassan)

See also: Qwen 2.5-Max outperforms DeepSeek V3 in some benchmarks

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Microsoft and OpenAI probe alleged data theft by DeepSeek appeared first on AI News.

The Chinese firm’s chat app, which recently soared to the top of Apple’s App Store, issued a notice on its website stating that only users with China-based phone numbers (+86) would be permitted to register for the foreseeable future.  

In the notice, DeepSeek said: “DeepSeek’s online services have recently faced large-scale malicious attacks. To ensure continued service, registration is temporarily limited to +86 phone numbers. Existing users can log in as usual.”

As of writing, sign-ups have reopened with a warning that “registration may be busy” and to please wait and try again.

Despite the challenges, existing users remain unaffected in terms of access. DeepSeek’s status page reported ongoing issues on Monday morning, adding to a turbulent day for the breakthrough AI platform.

DeepSeek faces test after explosive growth

DeepSeek has quickly become a disruptive force in the race for AI supremacy. In addition to gaining widespread user adoption, the app’s powerful open-source AI model has sparked waves of concern among US-based Big Tech firms.

The app’s strength lies in its ability to deliver robust AI performance on less-advanced chips, creating a more cost-effective and accessible solution compared to high-profile rivals such as OpenAI’s ChatGPT.  

This key differentiation shook investor confidence on Monday, leading to a significant dip in share prices for major US technology firms. Industry watchers suggest that such shocks may become more frequent as innovative competitors like DeepSeek challenge the dominance of traditional tech players.

Nigel Green, CEO of global financial advisory firm deVere Group, commented: “China’s technological advances, particularly in AI, are eroding the US’s ability to use tariffs as a tool to maintain global supremacy.

“The balance of power is shifting, and Washington must recognise that it can’t always dictate terms to Beijing as it once did. This new reality will have far-reaching consequences for investors and policymakers.

“AI will be the defining technology of our time, and the race to dominate it will shape global markets for decades to come.”  

Green went on to highlight the potential for non-traditional players and regions to seize the AI crown.

“Traditional tech giants are no longer the guaranteed winners. The focus must now shift to sectors and regions that are driving the next wave of innovation. This includes not only AI but also the critical infrastructure needed to support and secure it.”

The latest developments come against the broader canvas of growing competition between China and the US in the domain of AI and emerging technologies.

Despite export restrictions on US hardware, Beijing’s AI sector has been powering ahead—spurred by significant government investments and a burgeoning ecosystem of innovative startups.  

The rise of apps like DeepSeek signals that the playing field is no longer tilted decisively in favour of Silicon Valley. For years, US companies have been buoyed by leading semiconductor technologies from domestic firms like NVIDIA.

China’s ability to deliver competitive AI capabilities on less-advanced hardware could mark a significant turning point. However, while AI innovation is ramping up globally, DeepSeek’s struggles highlight the growing pains that can accompany explosive growth.

(Photo by Markus Spiske)

See also: Yiannis Antoniou, Lab49: OpenAI Operator kickstarts era of browser AI agents

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek restricts sign-ups amid ‘large-scale malicious attacks’ appeared first on AI News.

DeepSeek-R1-Zero is trained solely through large-scale reinforcement learning (RL) without relying on supervised fine-tuning (SFT) as a preliminary step. According to DeepSeek, this approach has led to the natural emergence of “numerous powerful and interesting reasoning behaviours,” including self-verification, reflection, and the generation of extensive chains of thought (CoT).

“Notably, [DeepSeek-R1-Zero] is the first open research to validate that reasoning capabilities of LLMs can be incentivised purely through RL, without the need for SFT,” DeepSeek researchers explained. This milestone not only underscores the model’s innovative foundations but also paves the way for RL-focused advancements in reasoning AI.

However, DeepSeek-R1-Zero’s capabilities come with certain limitations. Key challenges include “endless repetition, poor readability, and language mixing,” which could pose significant hurdles in real-world applications. To address these shortcomings, DeepSeek developed its flagship model: DeepSeek-R1.

Introducing DeepSeek-R1

DeepSeek-R1 builds upon its predecessor by incorporating cold-start data prior to RL training. This additional pre-training step enhances the model’s reasoning capabilities and resolves many of the limitations noted in DeepSeek-R1-Zero.

Notably, DeepSeek-R1 achieves performance comparable to OpenAI’s much-lauded o1 system across mathematics, coding, and general reasoning tasks, cementing its place as a leading competitor.

DeepSeek has chosen to open-source both DeepSeek-R1-Zero and DeepSeek-R1 along with six smaller distilled models. Among these, DeepSeek-R1-Distill-Qwen-32B has demonstrated exceptional results—even outperforming OpenAI’s o1-mini across multiple benchmarks.

• MATH-500 (Pass@1): DeepSeek-R1 achieved 97.3%, eclipsing OpenAI (96.4%) and other key competitors.  
• LiveCodeBench (Pass@1-COT): The distilled version DeepSeek-R1-Distill-Qwen-32B scored 57.2%, a standout performance among smaller models.  
• AIME 2024 (Pass@1): DeepSeek-R1 achieved 79.8%, setting an impressive standard in mathematical problem-solving.

A pipeline to benefit the wider industry

DeepSeek has shared insights into its rigorous pipeline for reasoning model development, which integrates a combination of supervised fine-tuning and reinforcement learning.

According to the company, the process involves two SFT stages to establish the foundational reasoning and non-reasoning abilities, as well as two RL stages tailored for discovering advanced reasoning patterns and aligning these capabilities with human preferences.

“We believe the pipeline will benefit the industry by creating better models,” DeepSeek remarked, alluding to the potential of their methodology to inspire future advancements across the AI sector.

One standout achievement of their RL-focused approach is the ability of DeepSeek-R1-Zero to execute intricate reasoning patterns without prior human instruction—a first for the open-source AI research community.

Importance of distillation

DeepSeek researchers also highlighted the importance of distillation—the process of transferring reasoning abilities from larger models to smaller, more efficient ones, a strategy that has unlocked performance gains even for smaller configurations.

Smaller distilled iterations of DeepSeek-R1 – such as the 1.5B, 7B, and 14B versions – were able to hold their own in niche applications. The distilled models can outperform results achieved via RL training on models of comparable sizes.

Bonus: Open-Source Distilled Models!

Distilled from DeepSeek-R1, 6 small models fully open-sourced
32B & 70B models on par with OpenAI-o1-mini
Empowering the open-source community

Pushing the boundaries of **open AI**!

2/n pic.twitter.com/tfXLM2xtZZ

— DeepSeek (@deepseek_ai) January 20, 2025

For researchers, these distilled models are available in configurations spanning from 1.5 billion to 70 billion parameters, supporting Qwen2.5 and Llama3 architectures. This flexibility empowers versatile usage across a wide range of tasks, from coding to natural language understanding.

DeepSeek has adopted the MIT License for its repository and weights, extending permissions for commercial use and downstream modifications. Derivative works, such as using DeepSeek-R1 to train other large language models (LLMs), are permitted. However, users of specific distilled models should ensure compliance with the licences of the original base models, such as Apache 2.0 and Llama3 licences.

(Photo by Prateek Katyal)

See also: Microsoft advances materials discovery with MatterGen

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post DeepSeek-R1 reasoning models rival OpenAI in performance  appeared first on AI News.