
Just as there are widely understood empirical laws of nature - for example, what goes up must come down, or every action has an equal and opposite reaction - the field of AI was long defined by a single idea: that more compute, more training data and more parameters makes a better AI model.
However, AI has since grown to need three distinct laws that describe how applying compute resources in different ways impacts model performance. Together, these AI scaling laws - pretraining scaling, post-training scaling and test-time scaling, also called long thinking - reflect how the field has evolved with techniques to use additional compute in a wide variety of increasingly complex AI use cases.
The recent rise of test-time scaling - applying more compute at inference time to improve accuracy - has enabled AI reasoning models, a new class of large language models (LLMs) that perform multiple inference passes to work through complex problems, while describing the steps required to solve a task. Test-time scaling requires intensive amounts of computational resources to support AI reasoning, which will drive further demand for accelerated computing.
What Is Pretraining Scaling? Pretraining scaling is the original law of AI development. It demonstrated that by increasing training dataset size, model parameter count and computational resources, developers could expect predictable improvements in model intelligence and accuracy.
Each of these three elements - data, model size, compute - is interrelated. Per the pretraining scaling law, outlined in this research paper, when larger models are fed with more data, the overall performance of the models improves. To make this feasible, developers must scale up their compute - creating the need for powerful accelerated computing resources to run those larger training workloads.
This principle of pretraining scaling led to large models that achieved groundbreaking capabilities. It also spurred major innovations in model architecture, including the rise of billion- and trillion-parameter transformer models, mixture of experts models and new distributed training techniques - all demanding significant compute.
And the relevance of the pretraining scaling law continues - as humans continue to produce growing amounts of multimodal data, this trove of text, images, audio, video and sensor information will be used to train powerful future AI models.
Pretraining scaling is the foundational principle of AI development, linking the size of models, datasets and compute to AI gains. Mixture of experts, depicted above, is a popular model architecture for AI training. What Is Post-Training Scaling? Pretraining a large foundation model isn't for everyone - it takes significant investment, skilled experts and datasets. But once an organization pretrains and releases a model, they lower the barrier to AI adoption by enabling others to use their pretrained model as a foundation to adapt for their own applications.
This post-training process drives additional cumulative demand for accelerated computing across enterprises and the broader developer community. Popular open-source models can have hundreds or thousands of derivative models, trained across numerous domains.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Post-training techniques can further improve a model's specificity and relevance for an organization's desired use case. While pretraining is like sending an AI model to school to learn foundational skills, post-training enhances the model with skills applicable to its intended job. An LLM, for example, could be post-trained to tackle a task like sentiment analysis or translation - or understand the jargon of a specific domain, like healthcare or law.
The post-training scaling law posits that a pretrained model's performance can further improve - in computational efficiency, accuracy or domain specificity - using techniques including fine-tuning, pruning, quantization, distillation, reinforcement learning and synthetic data augmentation.
Fine-tuning uses additional training data to tailor an AI model for specific domains and applications. This can be done using an organization's internal datasets, or with pairs of sample model input and outputs.
Distillation requires a pair of AI models: a large, complex teacher model and a lightweight student model. In the most common distillation technique, called offline distillation, the student model learns to mimic the outputs of a pretrained teacher model.
Reinforcement learning, or RL, is a machine learning technique that uses a reward model to train an agent to make decisions that align with a specific use case. The agent aims to make decisions that maximize cumulative rewards over time as it interacts with an environment - for example, a chatbot LLM that is positively reinforced by thumbs up reactions from users. This technique is known as reinforcement learning from human feedback (RLHF). Another, newer technique, reinforcement learning from AI feedback (RLAIF), instead uses feedback from AI models to guide the learning process, streamlining post-training efforts.
Best-of-n sampling generates multiple outputs from a language model and selects the one with the highest reward score based on a reward model. It's often used to improve an AI's outputs without modifying model parameters, offering an alternative to fine-tuning with reinforcement learning.
Search methods explore a range of potential decision paths before selecting a final output. This post-training technique can iteratively improve the model's responses
Most recent headlines
05/01/2027
Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be ...
07/10/2026
Dalet, a leading technology and service provider for media-rich organizations, today announced the latest Long-Term Supported (LTS) release of Dalet Flex. Build...
06/09/2026
June 9 2026, 23:00 (PDT) Dolby and MagentaTV Bring Fans Closer to the FIFA Worl...
06/08/2026
The broadcaster's 14-race schedule begins in Iowa and ends in November with ...
06/08/2026
In-venue and creative video staffers at the professional and collegiate level ha...
06/08/2026
As media organizations consolidate teams, technology platforms, and content libr...
06/08/2026
Firmware updates introduce Dolby's next-generation picture engine, content-aware optimization, and new motion and ambient-light tools...
06/08/2026
TNT Sports will have wall-to-wall coverage of the upcoming FIBA Women's Bask...
06/08/2026
swXtch.io (Stand 5.MR13) has unveiled groundSwXtch for Audio over IP (AoIP), a software application that lets broadcasters and media organizations move audio in...
06/08/2026
At IBC2026, Riedel Communications (Stands: 10.A24, 10.A31, 10.A38) will bring modern media production to life with the Live Production Experience - a connected ...
06/08/2026
The first live-sports director to receive the prestigious honor, eight-time Emmy...
06/08/2026
NBC Sports will continue to present the Preakness Stakes on NBC and Peacock thro...
06/08/2026
Cobalt Digital (Stand 8.F90) continues to reinforce its commitment to delivering...
06/08/2026
Wave Central has promoted promotion Jeff Daubert to Director of Sales, Americas.
In his expanded role, Daubert will lead Wave Central's sales strategy and ...
06/08/2026
KMH Integration is expanding its ability to serve media, broadcast, and IT organizations by adding Keith Hanadel, an accomplished architect with decades of indu...
06/08/2026
Advanced Systems Group (ASG) has promoted Michele Ferreira from Vice President of its Systems Integration team to the newly created position of Chief Business O...
06/08/2026
Demonstrations will include new JPEG XS, Dolby audio, cloud-gateway, stream-processing, and distribution capabilities...
06/08/2026
The deal includes three premium mobile units, engineering personnel, existing br...
06/08/2026
Broadcast Management Group produced Amazon Prime's inaugural Obsessed Fest a...
06/08/2026
Exhibit will feature new variable-ND functionality, OLED viewfinders, remote-camera tools, and UHD production monitors...
06/08/2026
Compact full-frame E-mount lens targets sports, wildlife, and other long-range shooting applications...
06/08/2026
Live broadcast will honor Roger Federer and Mary Carillo on Aug. 29, with onsite...
06/08/2026
End-zone board receives higher-resolution LED technology and an updated Show Control system ahead of the 2026 football season...
06/08/2026
New 16- and 32-pad MIDI controllers now available
Fender have just announced that the two new performance-focused controllers that were unveiled at NAMM 202...
06/08/2026
Standalone version, performance improvements & more
Lewitt have just released a significant update for their speaker and room virtualisation software, intro...
06/08/2026
Combines celeste samples with synth engine
After launching earlier in 2026 with deFORM, ebbandflow have returned with their first sampled acoustic instrumen...
06/08/2026
SGL Carbon's consolidated sales in the first half of 2026 totaled 394.2 million, down 13.0% from the prior year (H1 2025: 453.2 million). On the one hand,...
06/08/2026
The BBC has announced casting for the forthcoming six-part series Hercule from Mammoth Screen and Agatha Christie Limited.
Joining Bluemel are Henry Ashton (A ...
06/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
06/08/2026
Proven offering builds on more than 20,000 live sporting events with greater automation, flexibility, and operational control
LTN , the industry leader in tran...
06/08/2026
Hornets Tech, a specialist in broadcast connectivity and IP video, is bringing a new range of encoders, decoders and an IPTV platform to IBC2026.
Broadcast in...
06/08/2026
Introducing Adobe for ChatGPT: Create, edit and get work done - all in ChatGPT
Deepti Pradeep August 6, 2026
0 Comments
The Adobe plugin in ChatGPT br...
06/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
06/08/2026
At IBC2026, MediaKind will make its first major appearance as a unified global powerhouse in video, showcasing one of the world's most comprehensive video i...
06/08/2026
Big Blue Marble (#5.A63) will demonstrate how its integrated technology and operational expertise help media companies scale premium services with less complexi...
06/08/2026
LA JOLLA, CA-Scripps Research has received more than $500,000 in first-year fund...
06/08/2026
LA JOLLA, CA-Viruses are masters at invading our cells thanks to specialized proteins that coat their surfaces. When scientists design vaccines, they often crea...
06/08/2026
LA JOLLA, CA-The brain has its own immune system, which detects threats and mounts a defense. A growing body of evidence has shown that in Alzheimer's disea...
06/08/2026
LA JOLLA, CA-Scripps Research chemist Jin-Quan Yu has been elected to the National Academy of Sciences (NAS), one of the highest honors a scientist can achieve....
06/08/2026
LA JOLLA, CA-Scripps Research ranked third in the inaugural 2026 Cure Innovation Index recognizing the top-performing institutes and centers across the United S...
06/08/2026
Benjamin Cravatt, the Gilula Chair of Chemical Biology and a professor of chemis...
06/08/2026
LA JOLLA, CA-Dennis Burton, professor and the James & Jessie Minor Chair in Immu...
06/08/2026
LA JOLLA, CA-Inside every human cell, proteins are constantly being tagged with small chemical modifications after they're produced. Known as post-translati...
06/08/2026
LA JOLLA-Scripps Research has established the Ian Wilson Endowed Chair, a new fa...
06/08/2026
Scripps Research's Skaggs Graduate School of Chemical and Biological Science...
06/08/2026
LA JOLLA, CA-Professor Jin-Quan Yu of Scripps Research has been elected to the Fellowship of the Royal Society, the U.K.'s national academy of sciences and ...
06/08/2026
LA JOLLA, CA-For years, researchers have been hoping for vaccines that protect people against not just one strain of HIV, but every strain of the quickly mutati...
06/08/2026
LA JOLLA, CA-The Calibr-Skaggs Institute for Innovative Medicines, the nonprofit...
06/08/2026
LA JOLLA, CA-Building the complex 3D molecules needed for new medicines has alwa...
06/08/2026
LA JOLLA, CA-Fentanyl and related variants of the synthetic opioid kill more Americans each year than car accidents and gun violence combined. In too-high doses...