AI & ML news: Week 7 -13 October. AI get 2 nobel prizes, OpenAI’s DevDay… | by Salvatore Raieli | Oct, 2024

WEEKLY AI NEWS: RESEARCH, NEWS, RESOURCES, AND PERSPECTIVES

Probably the most attention-grabbing information, repository, articles, and assets of the week

Examine and star this repository the place the information will likely be collected and listed:

You will see the information first in GitHub. All of the Weekly Information tales are additionally collected right here:

• A multimodal generative AI copilot for human pathology. PathChat is a vision-language AI assistant designed for pathology, combining a foundational imaginative and prescient encoder and a big language mannequin, attaining state-of-the-art efficiency on diagnostic duties and outperforming different multimodal AI techniques, with potential purposes in training, analysis, and medical decision-making.
• Meta Movie Gen. Meta has developed a cutting-edge film mannequin with 30 billion parameters, which required 6,144 H100 GPUs for coaching. The mannequin was skilled utilizing 1 billion pictures and 100 million fastidiously chosen movies. Notably, it’s based mostly on a Temporal Autoencoder and incorporates Circulate matching Llama. Meta additionally printed a extremely detailed 92-page analysis paper, making it one of the vital complete experiences on the topic.
• When a language model is optimized for reasoning, does it still show embers of autoregression? An analysis of OpenAI o1. Massive language fashions face limitations as a result of they depend on subsequent token prediction. Though OpenAI’s o1 mannequin was skilled with a brand new goal centered on reasoning traces, it nonetheless reveals a few of the similar constraints related to subsequent token prediction.
• Contextual Document Embeddings. This paper presents a way just like a impartial TF/IDF, because it gathers info from your entire corpus fairly than counting on particular person doc embeddings. It successfully captures contextual info from surrounding paperwork and has achieved state-of-the-art outcomes on the MTEB benchmark.
• PairDistill: Pairwise Relevance Distillation for Dense Retrieval. This undertaking introduces a novel approach known as Pairwise Relevance Distillation (PairDistill), geared toward enhancing the accuracy of dense retrieval strategies.
• Modeling relationships to solve complex problems efficiently. Affiliate Professor Julian Shun develops high-performance algorithms and frameworks for large-scale graph processing.
• Factual Accuracy in AI. Integrative Decoding is a way designed to enhance the factual accuracy of huge language fashions, significantly for open-ended duties. This methodology helps guarantee extra dependable and correct outputs by refining the mannequin’s skill to combine info throughout technology.
• Dynamic Diffusion Transformer. The Dynamic Diffusion Transformer (DyDiT) improves the effectivity of diffusion fashions in picture technology by constructing on the Diffusion Transformer (DiT). It achieves this by dynamically adjusting computational assets throughout totally different timesteps and spatial areas, minimizing redundancy and optimizing efficiency.
• Redefining Temporal Modeling in Video Diffusion: The Vectorized Timestep Approach. The Body-Conscious Video Diffusion Mannequin (FVDM) enhances video technology by overcoming the restrictions of present fashions. As a substitute of utilizing a single timestep for your entire video clip, FVDM introduces a vectorized timestep variable, enabling every body to observe its personal noise schedule. This strategy improves the standard and coherence of generated movies.
• What Matters for Model Merging at Scale? Mannequin merging is a way that enables the mixture of two fashions to realize the efficiency advantages of each. Nonetheless, it doesn’t all the time scale successfully with bigger mannequin sizes. This paper investigates the necessities and challenges for making mannequin merging work effectively with very massive fashions, addressing points associated to scalability, efficiency trade-offs, and optimum merging methods.
• nGPT: Normalized Transformer with Representation Learning on the Hypersphere. A major quantity of analysis effort is targeted on normalizing the interior representations of language fashions. This examine demonstrates that by inserting each inside vector on a hypersphere, convergence time is considerably lowered for fashions of cheap measurement, resulting in extra environment friendly coaching.
• Genomic Foundation Model Benchmarking. GFMBench is a newly developed framework geared toward tackling challenges within the growth of genomic basis fashions (GFMs) by providing standardized benchmarking instruments. It helps the analysis of GFMs with tens of millions of genomic sequences and lots of of duties, automating the benchmarking course of for open-source GFMs to streamline their growth and comparability.
• LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations. This examine offers additional proof that language fashions internally encode indicators after they produce non-factual info. Understanding these inside cues may also help information fashions extra successfully and cut back the incidence of hallucinations, providing a possible technique for enhancing their reliability.
• Differential Transformer. Transformers usually over-allocate consideration to irrelevant context, resulting in inefficiencies. This analysis presents the Diff Transformer, which reinforces consideration to related info whereas filtering out noise. It introduces a differential consideration mechanism that computes consideration scores by subtracting two separate softmax consideration maps. This subtraction successfully cancels out noise and encourages sparse, extra centered consideration patterns, enhancing the mannequin’s efficiency on duties requiring exact context understanding.

• Brave New World: Leo AI and Ollama Bring RTX-Accelerated Local LLMs to Brave Browser Users. Nvidia’s RTX-Acceleration mixed with Ollama permits for operating native fashions within the browser.
• Liquid Foundation Models. Liquid AI has launched its first technology of Liquid Basis Fashions (LFMs), providing state-of-the-art efficiency whereas minimizing reminiscence consumption. The LFMs, that are optimized for various {hardware} platforms, embody 1B, 3B, and 40B parameter fashions. These fashions are already accessible on platforms like LIQUID PLAYGROUND and can quickly be out there on Cerebras. They’re significantly adept at processing sequential information and supply improvements in effectivity and scalability throughout industries like monetary providers and biotechnology.
• Introducing Copilot Labs and Copilot Vision. Microsoft is launching Copilot Labs to check superior AI instruments, together with Suppose Deeper and Copilot Imaginative and prescient. These instruments goal to develop the capabilities of their AI techniques, providing enhanced performance and deeper insights.
• OpenAI’s DevDay brings Realtime API and other treats for AI app developers. It’s been a tumultuous week for OpenAI, filled with govt departures and main fundraising developments, however the startup is again at it, making an attempt to persuade builders to construct instruments with its AI fashions at its 2024 DevDay. The corporate introduced a number of new instruments Tuesday, together with a public beta of its “Realtime API”, for constructing apps with low-latency, AI-generated voice responses. It’s not fairly ChatGPT’s Superior Voice Mode, however it’s shut.
• Microsoft brings AI-powered overviews to Bing. Microsoft has launched Bing generative search, an AI-driven function that gathers and summarizes info from the online, providing customers extra concise and aggregated search outcomes.
• KoBold Metals, which uses AI to help find critical minerals for the energy transition, raises $491M. Earlier this yr, KoBold Metals discovered what is likely to be one of many largest high-grade copper deposits of all time, with the potential to supply lots of of 1000’s of metric tons per yr, the corporate’s CEO mentioned.
• OpenAI gets $4 billion revolving credit line, giving it more than $10 billion in liquidity. OpenAI has secured over $10 billion in liquidity, attaining a valuation of $157 billion following its newest funding spherical. The corporate raised $6.6 billion from key buyers, together with Microsoft and Nvidia, however is contending with substantial operational prices, significantly the necessity for added GPUs to assist massive language mannequin (LLM) coaching. OpenAI is at the moment exploring restructuring methods to reinforce monetary progress and sustainability throughout the AI business.
• Black Forest Labs, the startup behind Grok’s image generator, releases an API. Black Forest Labs, the Andreessen Horowitz-backed startup behind the picture technology element of xAI’s Grok assistant, has launched an API in beta — and launched a brand new mannequin.
• DataPelago raises $47M to optimize hardware for analytical workloads. LLMs rely upon huge quantities of unstructured information for coaching, however this information requires in depth cleansing and processing earlier than it turns into helpful. Conventional information processing techniques, that are based mostly on CPUs and present software program architectures, weren’t designed to deal with the size and complexity of such information, leading to sluggish and expensive information preparation that hinders AI growth. To deal with these challenges, DataPelago has launched a Common Information Processing Engine, designed to beat efficiency, value, and scalability limitations, making AI growth sooner and extra inexpensive.
• Google brings ads to AI Overviews as it expands AI’s role in search. Google will start to indicate advertisements in AI Overviews, the AI-generated summaries it provides for sure Google Search queries, and can add hyperlinks to related net pages for a few of these summaries as nicely. It’s additionally rolling out AI-organized search outcomes pages within the U.S. this week.
• Nobel Physics Prize Awarded for Pioneering A.I. Research by 2 Scientists. Two scientists who contributed to the event of neural networks have been awarded the Nobel Prize in Physics, recognizing their groundbreaking work in advancing synthetic intelligence and neural community applied sciences.
• Introducing the Message Batches API. Anthropic has launched a brand new batch processing API that enables builders to submit batches of as much as 10,000 queries without delay. Every batch is processed inside 24 hours and is 50% cheaper than customary API calls, making it a extra environment friendly and cost-effective resolution for dealing with non-time-sensitive duties.
• Update on Reflection-70B. An in depth autopsy evaluation of the extremely anticipated Reflection-70B mannequin revealed points with its benchmark code, which inflated its efficiency claims. Though the staff has since corrected these bugs, and the mannequin’s efficiency stays spectacular, it doesn’t fairly attain the initially marketed ranges.
• Four-legged robot learns to climb ladders. The proliferation of robots like Boston Dynamics’ Spot has showcased the flexibility of quadrupeds. These techniques have thrived at strolling up stairs, traversing small obstacles, and navigating uneven terrain. Ladders, nevertheless, nonetheless current a giant difficulty — particularly given how ever current they’re in factories and different industrial environments the place the techniques are deployed.
• Braintrust raises $36M Series A. Braintrust, which helps Airtable, Brex, Notion, and Stripe construct AI merchandise, has raised $36M in a Sequence A led by a16z.
• Clout Kitchen raises $4.45M for AI gaming pal that mimics content creators. Clout Kitchen introduced in the present day that it has raised $4.45 million in its seed funding spherical, which it plans to place in the direction of its new creator-powered merchandise and experiences. The primary of those is Backseat AI, an AI-powered buddy for League of Legends that the corporate created with Tyler “Tyler1” Steinkamp — an AI buddy that may tackle the facet of fashionable gaming content material creators. Clout Kitchen plans to make use of its funding to develop its staff and construct out its shared inside tech stack.
• AlphaFold wins Nobel Prize in Chemistry. Demis Hassabis, John Jumper, and David Baker had been awarded the Nobel Prize in Chemistry for his or her groundbreaking work in protein folding, significantly by means of improvements like AlphaFold. Their contributions have considerably superior the understanding of protein constructions and their implications for science and medication.
• OpenAI reducing dependency on Microsoft data centers. OpenAI is reducing its reliance on Microsoft’s information facilities by buying its personal compute infrastructure, permitting larger independence in its operations. Concurrently, Microsoft is decreasing its dependence on OpenAI because it develops and competes with its personal AI merchandise, signaling a shift within the dynamics of their partnership.
• TikTok parent company ByteDance has a tool that’s scraping the web 25 times faster than OpenAI. TikTok dad or mum firm ByteDance is amassing big volumes of net information means sooner than the opposite main net crawlers. ByteDance could also be planning to launch its personal LLM, and is aggressively utilizing its net crawler, “Bytespider,” to scrape up information to coach its fashions, Fortune reported.
• Sonair takes a cue from dolphins to build autonomous 3D vision without lidar. Ultrasound is probably finest often known as the expertise that permits noninvasive physique scans and underwater communication and may also help us park our automobiles. A younger startup known as Sonair out of Norway desires to make use of it for one thing else: 3D pc imaginative and prescient utilized in autonomous {hardware} purposes.
• Tesla’s head of vehicle programs jumps to Waymo ahead of robotaxi reveal. Tesla has misplaced a prime govt to Waymo within the lead-up to the EV maker’s robotaxi unveiling on Thursday.
• Autism ABA Therapy with Llama. Meta shares a use case of its Llama mannequin for medical and therapeutic profit.
• Uber’s EV ridehailing business is maturing. The corporate additionally introduced it was including ChatGPT to its driver app to deal with EV questions.
• Amazon’s new AI guides can help shoppers find what they need. The brand new AI Buying Guides function goals to assist customers discover what they want with extra knowledgeable product strategies.
• TikTok joins the AI-driven advertising pack to compete with Meta for ad dollars. TikTok’s Sensible+ is an AI-powered ad-buying device designed to automate and optimize advert campaigns, giving entrepreneurs the choice to selectively make the most of its options for enhanced efficiency. The device seeks to rival Meta’s Benefit+ by providing streamlined advert administration and improved return on funding (ROI). Early outcomes point out vital positive factors in advert spend effectivity and conversion charges, positioning TikTok as a robust contender within the digital promoting market.
• OpenAI partners with Cosmopolitan and Elle publisher Hearst. ChatGPT will present citations and direct hyperlinks to the corporate’s content material.
• Meta debuts new generative AI tools for creating video-based ads. Meta Platforms Inc. in the present day mentioned it’s rolling out a full-screen video tab on Fb in recognition of the truth that its customers spend extra time watching movies than the rest on its platforms.

• Introducing the Open FinLLM Leaderboard. The Open FinLLM Leaderboard offers a devoted analysis platform designed particularly for monetary language fashions. It emphasizes key monetary duties like predicting inventory actions, analyzing sentiment, and extracting info from monetary experiences.
• Infinite-Fractal-Stream: Small Scale Proxy for Scaling-Centric ML. Mannequin testing within the picture area is usually constrained by low-quality, small datasets like CIFAR10. This GitHub repository offers a device that generates infinite, complicated fractals within the type of pictures or movies, providing a brand new strategy for testing fashions.
• Auto Jobs Applier. A extremely viral repository leverages language fashions to automate the job software course of, including an additional layer of personalization to tailor purposes for every place.
• Real-World Benchmarks Make Membership Inference Attacks Fail on Diffusion Models. This examine uncovers main weaknesses in present membership inference assaults (MIAs) used to detect unauthorized information utilization in diffusion fashions. It introduces CopyMark, a extra sensible benchmark for assessing MIAs on pre-trained fashions, offering unbiased datasets and truthful analysis strategies to enhance the accuracy and reliability of those assaults.
• ImageFolder: Autoregressive Image Generation with Folded Tokens. ImageFolder is a semantic tokenizer developed to stability the trade-off between picture reconstruction accuracy and technology high quality in visible generative fashions, enhancing the general efficiency of those fashions in each duties.
• Grounded-VideoLLM: Sharpening Fine-grained Temporal Grounding in Video Large Language Models. Grounded-VideoLLM is a novel Video-Massive Language Mannequin (Video-LLM) created to reinforce the fine-grained understanding of particular moments in movies. By incorporating a temporal stream and discrete temporal tokens, the mannequin extra successfully captures the relationships between frames and timestamps, enhancing its skill to interpret and analyze detailed video content material.
• Autoregressive Action Sequence Learning for Robotic Manipulation. The Chunking Causal Transformer (CCT) is a brand new autoregressive structure developed particularly for robotic manipulation duties. It’s designed to enhance the mannequin’s skill to course of sequential information effectively, optimizing efficiency in real-time robotic management and manipulation situations.
• FacePoke. FacePoke is a device designed for fast enhancing of faces in each movies and pictures, permitting customers to make fast changes and modifications with ease.
• pipeline_parallel.py. A big mannequin coaching lead at Hugging Face has shared a superb 200-line instance of parallelism constructed from scratch, demonstrating environment friendly strategies for distributing computational duties, which is especially helpful for large-scale mannequin coaching.
• CodeMMLU: A Multi-Task Benchmark for Assessing Code Understanding Capabilities of CodeLLMs. As language fashions change into more and more proficient at writing code, many present benchmarks are approaching saturation. This paper proposes a tougher benchmark designed to evaluate how nicely fashions carry out on reasoning and code technology duties, pushing past fundamental code-writing capabilities to guage deeper problem-solving abilities.Intensify.Intensify is a Python package deal that lets you colorize textual content based mostly on depth values. It offers an easy-to-use interface for making use of shade gradients to textual content or background colours within the terminal.
• Beyond FVD: Enhanced Evaluation Metrics for Video Generation Quality. JEDi is a brand new metric constructed on the Joint Embedding Predictive Structure (JEPA), designed to reinforce analysis accuracy with fewer samples. It higher aligns with human assessments, making it a extra strong various to the FVD (Fréchet Video Distance) metric for evaluating generative fashions.
• PRFusion: Toward Effective and Robust Multi-Modal Place Recognition with Image and Point Cloud Fusion. PRFusion and PRFusion++ are multimodal fashions developed to reinforce place recognition in robotics and pc imaginative and prescient. By combining info from a number of sensory inputs, these fashions enhance the accuracy and robustness of place recognition duties, making them more practical in real-world purposes.
• Fine-Tuning CLIP’s Last Visual Projector: A Few-Shot Cornucopia. This paper presents ProLIP, a novel methodology for adapting vision-language fashions resembling CLIP with out including further parameters. ProLIP fine-tunes solely the ultimate projection matrix of the imaginative and prescient encoder, enabling it to ship sturdy efficiency in few-shot classification duties whereas sustaining the mannequin’s effectivity.
• ScienceAgentBench. The benchmark code for the science agent check is designed to guage how successfully fashions can contribute to novel scientific discoveries. It offers a framework for assessing a mannequin’s skill to generate modern concepts, clear up complicated scientific issues, and make significant advances in numerous scientific fields.
• Controlled Visual Generation. Controllable AutoRegressive Modeling (CAR) is a novel framework that introduces exact management mechanisms to pre-trained visible autoregressive fashions. This methodology permits extra refined and focused picture technology by progressively enhancing management representations, permitting for fine-tuned outputs with lowered computational assets.
• PredFormer: Transformers Are Effective Spatial-Temporal Predictive Learners. PredFormer is a newly developed transformer-based methodology for spatiotemporal predictive studying, providing superior efficiency in each accuracy and effectivity in comparison with present approaches. It excels in duties that contain predicting adjustments over time and area, making it a robust device for numerous purposes in fields like video evaluation, climate forecasting, and robotics.
• GenSim2: Scaling Robotic Data Generation with Multi-modal and Reasoning LLMs. This paper presents an modern strategy to scaling robotic information assortment by using an enhanced, high-quality physics simulation dataset. The improved simulation surroundings permits extra environment friendly information technology for coaching robots, providing a scalable and cost-effective methodology to gather massive quantities of correct and numerous information for robotic studying and growth.
• Learning Efficient and Effective Trajectories for Differential Equation-based Image Restoration. This undertaking introduces a novel differential equation-based strategy for picture restoration. By leveraging mathematical fashions grounded in differential equations, the strategy enhances the power to recuperate and restore degraded or noisy pictures, offering improved accuracy and efficiency in picture restoration duties.
• Pixtral 12B. The Mistral staff has supplied detailed insights into the coaching course of and structure of their vision-language mannequin, which has demonstrated stable efficiency. The mannequin incorporates superior strategies for successfully integrating visible and linguistic information, permitting it to carry out nicely on a wide range of duties that require understanding each pictures and textual content. The shared info contains specifics on information preprocessing, mannequin structure, and the optimization methods employed throughout coaching.
• MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering. MLE-bench is a benchmark created to guage AI brokers’ capabilities in machine studying engineering. It features a curated collection of 75 Kaggle competitions to check numerous abilities, resembling mannequin coaching, dataset preparation, and optimization. The benchmark goals to evaluate how nicely AI brokers can deal with sensible machine studying duties, offering a complete analysis of their engineering proficiency.
• Deciphering Cross-Modal Alignment in Large Vision-Language Models with Modality Integration Rate. The Modality Integration Price (MIR) is a brand new metric designed to guage the effectiveness of multi-modal pre-training in Massive Imaginative and prescient Language Fashions. It measures how nicely totally different modalities, resembling visible and textual information, are built-in throughout the pre-training course of, providing insights into the mannequin’s skill to leverage info from each sources to enhance efficiency on multi-modal duties.
• Aria: First Open Multimodal Native MoE Model. A extremely spectacular new vision-language mannequin has been launched with open weights, code, and a complete analysis report. It achieves efficiency on par with closed fashions for lengthy video understanding, a problem that has confirmed troublesome for different open fashions like Pixtral and Molmo. This development represents a major breakthrough within the discipline of open-source vision-language fashions.
• IterComp: Iterative Composition-Aware Feedback Learning from Model Gallery for Text-to-Image Generation. IterComp is a brand new framework developed to reinforce compositional text-to-image technology by integrating the strengths of a number of superior diffusion fashions, together with RPG, Secure Diffusion 3, and FLUX. By leveraging these fashions, IterComp improves the standard and coherence of generated pictures, particularly when dealing with complicated textual prompts that require a number of parts to be composed precisely.
• MatMamba. MatMamba is a novel structure for sequence processing, constructing upon the Mamba2 framework by incorporating a Matryoshka-like design. This strategy permits a single mannequin to be skilled at a number of granularities, enabling the extraction of assorted smaller, nested submodels. This hierarchical construction enhances flexibility and effectivity, permitting the mannequin to adapt to totally different ranges of complexity and useful resource constraints.
• O1 replication progress report. Researchers from GAIR and NYU have been investigating the vital algorithmic developments behind OpenAI’s o1 mannequin’s distinctive efficiency. Of their report, they introduce the idea of “Journey Studying” information, a novel strategy that, when utilized in coaching, boosts math efficiency by 8% in absolute phrases. This innovation highlights how particular information varieties can considerably improve a mannequin’s reasoning and problem-solving talents.

• Nuclear power for AI: what it will take to reopen Three Mile Island safely. As Microsoft strikes a deal to restart a reactor on the infamous energy station, Nature talks to nuclear specialists concerning the unprecedented course of.
• ‘In awe’: scientists impressed by latest ChatGPT model o1. The chatbot excels at science, beating PhD students on a tough science check. However it would possibly ‘hallucinate’ greater than its predecessors.
• Can AI have common sense? Finding out will be key to achieving machine intelligence. The appearance of LLMs has reopened a debate concerning the limits of machine intelligence — and requires new benchmarks of what reasoning consists of.
• How your brain detects patterns in the everyday: without conscious thought. Neurons in sure mind areas combine ‘what’ and ‘when’ info to discern hidden order in occasions in actual time.
• AI to the rescue: how to enhance disaster early warnings with tech tools. Synthetic intelligence may also help to cut back the impacts of pure hazards, however strong worldwide requirements are wanted to make sure finest apply.
• Before Mira Murati’s surprise exit from OpenAI, staff grumbled its o1 model had been released prematurely. OpenAI’s accelerated growth and security testing of its newest fashions, resembling GPT-4o and o1, have led to inside friction, ensuing within the departure of a number of senior workers members. The fast tempo of growth has raised issues concerning the thoroughness of the protection protocols, contributing to tensions throughout the group.
• I Quit Teaching Because of ChatGPT. This professor resigned from educating as a result of widespread use of huge language fashions (LLMs) like ChatGPT amongst college students, which they felt undermined educational integrity and the standard studying course of.
• Three Subtle Examples of Data Leakage. This text examines the dangers of information leakage in machine studying, showcasing two real-world instances the place improper information dealing with resulted in deceptive mannequin efficiency. In a single occasion, an organization incorrectly filtered information by an higher worth restrict earlier than modeling, whereas one other group encountered issues by not following a strict chronological cut up. The important thing classes emphasize the vital want for detecting information leakage and understanding its detrimental results on mannequin accuracy and reliability.
• The real data wall is billions of years of evolution. AI growth is encountering a possible impediment often known as the “information wall,” as language fashions close to the restrict of accessible textual information for coaching. This text challenges the concept of utilizing human analogies to beat these information constraints, stating that human intelligence outcomes from huge quantities of information and lengthy evolutionary processes, which differ essentially from AI. Whereas human studying methods might in a roundabout way translate to AI, this doesn’t preclude progress by means of different modalities, resembling multimodal information, or developments in algorithms that would push AI capabilities additional.
• AI will use a lot of energy. That’s good for the climate. AI information facilities are considerably growing the demand for clear, 24/7 power, prompting tech giants to take a position closely in renewable and nuclear energy options. This rising demand is anticipated to speed up the associated fee discount of fresh power applied sciences, pushed by their studying charges. Over time, the power wants of AI may result in coverage shifts and developments in clear power infrastructure, fostering sooner adoption and growth of sustainable power sources.
• I want to break some laws too. This text explores the usage of an automatic information cleansing pipeline impressed by the Minipile methodology, which prunes datasets to ship vital efficiency positive factors with solely a fraction of the unique information measurement. By leveraging strategies resembling few-shot prompting and clustering, the strategy streamlines dataset refinement for AI coaching, difficult conventional scaling legal guidelines by prioritizing information high quality over amount. The outcomes point out that utilizing foundational datasets with extra refined information can optimize AI mannequin coaching, decreasing useful resource consumption whereas boosting efficiency.

What do you consider it? Some information that captured your consideration? Let me know within the feedback