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 ...
04/08/2026
Dalet, a leading technology and service provider for media-rich organizations, t...
04/07/2026
April 7 2026, 19:00 (PDT) Detective Conan: Fallen Angel of the Highway Opens in...
01/06/2026
January 6 2026, 05:30 (PST) Dolby Sets the New Standard for Premium Entertainment at CES 2026
Throughout the week, Dolby brings to life the latest innovatio...
18/05/2026
Despite a bumpy start to the year, the provider of event-production support and ...
18/05/2026
Daktronics has partnered with the University of North Carolina to manufacture and install 11 LED displays totaling more than 10,000 square feet and more than 14...
18/05/2026
CBS LA has announced a multi-year partnership with the Los Angeles Rams, covering exclusive local broadcasts of Rams preseason games, weekly year-round programm...
18/05/2026
Skyline Communications has announced an integration between its DataMiner xOps p...
18/05/2026
eCLUTCH, the hybrid esports platform powered by iKOMG, has announced an expansion of its distribution across Europe, MENA, Africa, and Asia, along with new cont...
18/05/2026
Behind The Mic provides a roundup of recent news regarding on-air talent, includ...
18/05/2026
With 22 games this season, the production team looks forward to tweaking and enhancing the coverage
After 8,660 days off the air, the WNBA returned to NBC yest...
18/05/2026
(L-R) Midori Francis and Natalie Erika James attend the Saccharine premiere during the 2026 Sundance Film Festival at The Ray Theatre on January 22, 2026, in ...
18/05/2026
Last night, the Spotify Podcast Awards in Mexico returned to the country's capital. Now in its second year, the evening honors creators whose voices are hel...
18/05/2026
ZEN-Core synth goes mobile
Roland's powerful ZEN-Core software synthesizer has just been introduced to the iPad, offering a convenient entry point into ...
18/05/2026
Versatile new limiter plug-in announced
Based in Sheffield, UK, fedDSP offer a range of plug-ins that span the music production, live sound and high-end med...
18/05/2026
A new use for convolution?
Viiri Audio's debut plug-in aims to do something a little different with convolution processing, allowing users to adjust all...
18/05/2026
Delta Goodrem shines for SBS as more than 3.27 million Australians tune in for E...
18/05/2026
The Australian Defence Force uses L3Harris T4 and T7 robots for explosive ordnan...
18/05/2026
Continued investment across Europe and Germany is expanding local teams and improving access to stock, regional expertise, and specialist broadcast support.
CV...
18/05/2026
18 May 2026
Dubai and New York, May 18, 2026 - VEON Ltd. (NASDAQ: VEON), a glob...
18/05/2026
Monday 18 May 2026
Sky News offers ad-free podcasts and bonus episodes for just...
18/05/2026
Comscore March 2026 Consumer AI Chatbot Usage Rankings Show Claude Gaining Share OpenAI's ChatGPT maintains lead while Anthropic's Claude continues to c...
17/05/2026
Delta Goodrem's Eurovision Eclipse marks end of a stellar run
17 May, 2026
Media releases
Bulgaria wins Eurovision 2026
Relive every spellbinding mome...
17/05/2026
Back to All News
Oasis Premieres on Netflix June 19
Entertainment
17 May 2026
GlobalSpain
Link copied to clipboard
Summer, sunshine, the beach, parties. T...
16/05/2026
Helps to deliver a clean, balanced midrange
Developed alongside Newfangled Audio, the latest plug-in in Eventide's software collection has been designed...
16/05/2026
Brings onboard stem rendering to RANE System One
Engine DJ have just released Engine DJ 5.0, a free update for their Engine DJ OS embedded hardware and Engi...
16/05/2026
Boris FX Continuum Pairs AI Precision and Advanced Creative Controls
Jessie Electa Petrov May 16, 2026
0 Comments
The 2026.5 release adds automatic de...
16/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
15/05/2026
Seattle Sounders FC and Seattle Reign FC, in partnership with RAVE Foundation an...
15/05/2026
Dan Brumm has served as sound designer on Bluey, the Australian children's t...
15/05/2026
The Professional Audio Manufacturers Alliance (PAMA) and Shure Incorporated are accepting applications for the 6th annual Mark Brunner Professional Audio Schola...
15/05/2026
Netflix has announced an expanded NFL schedule for 2026 and beyond under a four-year partnership extension with the NFL through the 2029-30 season. Each season,...
15/05/2026
Ateme is supporting TVRI (Televisi Republik Indonesia) with a contribution and d...
15/05/2026
Concacaf has announced the launch of a new website and mobile app built on Deltatre's FORGE platform. Concacaf.com and the mobile app, available on iOS and ...
15/05/2026
Eutelsat has announced the launch of QBC Business Economic Channel by Qatar Media Corporation, broadcasting in 4K/UHD via Eutelsat's 7/8 West video neighbo...
15/05/2026
Major League Soccer has announced four original content series timed to the 2026...
15/05/2026
The Alliance for IP Media Solutions (AIMS) has announced it will exhibit and present at InfoComm 2026, taking place June 13-19 at the Las Vegas Convention Cente...
15/05/2026
InfoComm 2026 will take place June 13-19 (exhibits June 17-19) at the Las Vegas Convention Center. The show will include sessions and exhibits covering broadcas...
15/05/2026
Tracy McGrady's Ones Basketball League (OBL) and FuboTV Inc. have announced ...
15/05/2026
Disguise has partnered with Creative Technology (CT) to deliver visual playback ...
15/05/2026
Sony Electronics has announced two new products for professional imaging: the Alpha 7R VI full-frame mirrorless camera and the FE 100-400mm F4.5 GM OSS super-te...
15/05/2026
In-venue and creative video staffers at the professional and collegiate level ha...
15/05/2026
Ratings Roundup is a rundown of recent rating news and is derived from press rel...
15/05/2026
For sports organizations, the most valuable assets are often the most sensitive:...
15/05/2026
The NFL's broadcast partners released their 2026 regular season schedules ye...
15/05/2026
Besides drawing on experience with boxing, the production integrates specialty c...
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.
