The creative and practical scope of LLMs is undergoing a dramatic expansion.
June 2025 - Vol. 68 No. 6
Features
Formal computer science may never model practical computer engineering well enough to be predictive in the way the physical sciences are.
Sutton and Barto developed reinforcement learning, a machine learning method that trains neural networks by offering them rewards in the form of numerical values.
Generated data that starts forgetting tail events can lead to a concentration of higher probability distributions, which causes a model to fail.
How AI measures up to the human motivation to experiment, break boundaries, or create something new.
Privacy, performance, and security benefits have everyone from academic computer scientists to technology giants racing to develop more efficient ways of pulling AI out of the cloud and closer to users.
We need people who understand not just technology, but also law, policy, ethics, and all in an international context.
In Large Language Models We Trust?
There may be no reliable way to “look up” the truth in real time via an Internet search for evidence to corroborate an LLM’s claim.
It turns out your mental state when writing code or solving a problem often has a bearing on the solution.
Shedding light on the future of programming in the age of generative AI.
Thoughts about Some Surprising AI-Era Technology Readiness Findings
Survey findings define the essence of risk in companies' business-technology plans and readiness.
Web 3.0 Requires Data Integrity
It’s time for new integrity-focused standards to enable the trusted AI services of tomorrow.
Systems Correctness Practices at Amazon Web Services
A survey of the portfolio of formal methods used across AWS to deliver complex services with high confidence in assuring systems correctness.
Revolutionizing Datacenter Networks via Reconfigurable Topologies
An overview of reconfigurable datacenter networks and their technological enablers.
From Prompt Engineering to Prompt Science with Humans in the Loop
Demonstrating how to have scientific rigor in developing a reliable prompt and getting a trustworthy response for a downstream application.
Malicious AI Models Undermine Software Supply-Chain Security
Trusted repositories, cryptographic validation, and controlled access can help organizations mitigate risks associated with malicious AI models.
The accompanying paper asks whether we can internalize in logic the property known as “Church’s Thesis.”
‘Upon This Quote I Will Build My Church Thesis’
The compatibility of the Church thesis with dependent type theory was an open question. In this paper, we answer this question positively.
Technical Perspective: A Symbolic Approach to Verifying Quantum Systems
The combination of deep theoretical insights and practical tool development results in a milestone for quantum circuit verification.
An Automata-Based Framework for Verification and Bug Hunting in Quantum Circuits
The framework leverages tree automata to compactly represent sets of quantum states; transformers implement the semantics of quantum gates over this representation.
Developing the Foundations of Reinforcement Learning
2024 Turing laureates Andrew G. Barto and Richard S. Sutton discuss the theoretical background and practical application of reinforcement learning.
