SShortSingh.
Back to feed

How Google Translate Turns Sentences Into Numbers to Cross 249 Languages

0
·23 views

Google Translate does not understand language the way people do; instead, it converts input text into numerical patterns and predicts likely translations by matching them against vast collections of human-made translations. Its neural system, which replaced an older phrase-based method in 2016 and cut errors sharply, can even translate between language pairs it was never directly trained on, creating its own internal intermediate representation. For unfamiliar words, a subword technique called WordPiece splits them into recognizable fragments such as "un" and "happiness". The article presents Google Translate as highly useful for everyday tasks while noting that it still makes errors that users should check.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Unpatched VMware vCenter zero-day exploited by Chinese APT group

A critical, unpatched remote code execution flaw in VMware vCenter was actively exploited in a zero-day campaign attributed to a sophisticated Chinese advanced persistent threat group. Because vCenter serves as the central management plane for virtualised infrastructure, a compromise could expose every system it manages, including databases, domain controllers, and backup systems. The vulnerability requires no valid credentials and can run commands with root or SYSTEM privileges, allowing attackers to potentially gain full control of a virtual environment. The article argues that such a flaw demands emergency out-of-band response rather than standard patch cycles, along with pre-authorised mitigation paths and detection controls. Recommended steps include applying vendor patches immediately, isolating hosts and rotating credentials if no fix exists, and restricting management-plane access to trusted administrative sources.

0
ProgrammingDEV Community ·

Physical AI Requires Hardware and Software Co-Design, Says STATERA Library Chapter

Physical AI systems cannot be treated as models separate from their physical platforms, since perception, computation, decisions, and actuation form one integrated loop. Hardware factors such as timing, energy efficiency, reliability, and thermal limits directly shape what an AI system can achieve. System requirements should be derived from the physical task and operating environment, with models and computing platforms chosen together to avoid impractical or overly constrained designs. Sensors define what the AI can observe, while actuators define the physical actions it can take, making their characteristics central to architecture. The chapter emphasizes that increasing capability brings unavoidable tradeoffs among performance, power, cost, weight, and bandwidth.

0
ProgrammingDEV Community ·

Physical AI Co-Design: Why Robot Hardware and AI Must Be Engineered Together

The chapter argues that intelligent robots cannot be designed well by treating the physical platform, software, and AI model as separate parts, since their capabilities are tightly linked to the body and environment. It holds that system performance comes from integration rather than optimizing any single component alone. The first principle is that the mission, task, and operating environment should shape the architecture, with requirements for latency, robustness, power, and size derived from what the system must accomplish. The text also treats the robot's physical embodiment as a central design variable, since geometry, mass, actuators, and sensors determine what the AI can perceive and how it can act. Finally, it stresses selecting AI models and hardware together, because more capable models demand more compute, memory, and power, and hardware-aware model design is needed for real-world deployment.

0
ProgrammingDEV Community ·

Compute Platforms for Physical AI: Sizing Hardware Around Real-Time Robotic Workloads

This chapter from the STATERA Library's Physical AI hardware-software co-design volume argues that compute requirements should be derived from the full system rather than from hardware specifications alone. It holds that a robot must continuously convert sensor data into understanding, decisions, and control actions under real-world constraints, so workloads like perception, planning, and control must be sized together. Latency from sensing to action, including data movement and control, is presented as more important than peak theoretical throughput. The text also explains that heterogeneous sensor streams and different processor types, such as CPUs, GPUs, NPUs, and MCUs, shape the architecture needed for these demands.