Arjun Singh

A November 2025 UC San Diego faculty report revealed that nearly one in eight incoming freshmen (11.8%) require remedial math courses covering middle school-level material—a thirtyfold increase since 2020. The University of California eliminated standardized test requirements in May 2020 citing equity concerns and racial bias, transitioning to "test-blind" admissions by 2023. However, the UC system has not reinstated testing requirements, even as elite universities like Harvard, Yale, MIT, Dartmouth, and Brown have reversed course and mandated SAT/ACT scores again. The crisis reflects compounding factors: #COVID19 learning loss, #gradeinflation, elimination of objective #admissionsmetrics, and increased enrollment from under-resourced schools—raising fundamental questions about academic preparedness and institutional mission. UC San Diego Report Exposes Thirtyfold Increase in Students Lacking Basic Math Skills as Elite Universities Nationwide Return to Testing Requirements December 6, 2025 The UC San Diego released a sobering faculty report in November revealing that remedial math enrollment has surged from under 1 percent to nearly 12 percent of incoming freshmen in just five years. The crisis has forced the prestigious research university to redesign courses to teach elementary and middle school #mathematics—making it the only UC campus offering credit-bearing courses for elementary-level math remediation. The report from UC San Diego's Senate-Administration Working Group on Admissions paints a troubling picture: between fall 2020 and fall 2025, students whose math placement exam results fell below middle school standards increased nearly thirtyfold. In fall 2024, 42 percent of students in the remedial Math 2 course had passed precalculus or calculus in high school, while 25 percent earned 4.0 GPAs in high school math—yet demonstrated proficiency below eighth-grade levels. "We are basically setting them up for failure," Professor Akos Rona-Tas, who helped author the report, told KPBS. "Many of these students would have been much better off going to say community college and taking the transfer route."

1

Podonos just raised $2.4 million in pre-seed funding, and the signal is loud and clear: the voice AI industry finally has someone measuring the sound, not just making the noise. Based in Los Gatos, Podonos is building the infrastructure layer that decides whether voice AI feels real, human, and ready for primetime. The industry is racing toward a projected $47.5 billion by 2034, but growth without standards is chaos. Podonos is making sure the hype actually holds up when you press play. At the center is ✦Soohyun Bae, PhD in Computer Engineering from Georgia Institute of Technology and Y Combinator W22 alumnus. His track record stretches from engineering roles at Google Maps to leading AR mapping at Niantic, plus co-founding Bobidi and TickTock AI. He has seen what happens when technology scales without proper #guardrails. With Podonos, he's betting the future of AI isn't just about how smart the models get, but how believable and trustworthy they sound. Podonos evaluates models across naturalness, similarity, emotion, recognition accuracy, pronunciation, tone, and resilience. Add in #personaconsistency, and suddenly you're not just testing machines, you're testing their ability to pass as human. The twist? They deliver results in under twelve hours, when the legacy approach drags on for two months. In AI time, that's the difference between relevance and irrelevance. Investors caught the beat. Serac Ventures, led by Kevin Moore, took the lead, backed by NAVER D2SF, the venture arm of South Korea's NAVER, and KAIST Venture Investment Holdings. Together, they've now backed Podonos twice, with this $2.4 million round stacking on top of a $750,000 raise in March 2025. Total funding sits at $3.15 million, and every dollar is pointed toward scaling #engineering, refining #analysistools, and launching into the U.S., Europe, and Southeast Asia. This isn't theory. Podonos already serves Resemble AI, Play AI, and Sanas AI, hitting six-digit ARR by November 2024 with just four employees. Their secret weapon is reach: 150,000 evaluators spread across nine languages and thirteen locales, forming a human-in-the-loop platform that blends speed, rigor, and global coverage. That reach lets Podonos deliver evaluations at a scale and pace that no competitor has matched. The strategy is ambitious: expand into high-demand sectors like #healthcare, #finance, #gaming, and #advertising, then move beyond voice into #multimodal model evaluation, spanning video AI and large language models. The mission isn't just to check if AI works, it's to decide if AI works for people. #Startups #StartupFunding #EarlyStage #VentureCapital #PreSeed #AI #AIInfrastructure #Voice #VoiceAI #VoiceTech #Infrastructure #Technology #Innovation #TechEcosystem #StartupEcosystem #Hiring #TechHiring If software engineering peace of mind is what you crave, Vention is your zen.

6

1 Comment

MACHINES INSTEAD OF HUMANS: The current U.S. job market feels extremely brutal for many new grads and experienced workers. A recent UC Davis grad applied to 200 jobs and was rejected 200 times; many businesses didn’t even respond, like the proverbial “F-off”. Millions feel frustrated. According to the article, “The Job Market is Hell”, by Annie Lowrey many feel frustrated with a low success rate. The current US jobless rate remains low (4.3%), even lower than the European Union average of 6%. However, hiring has slowed to its lowest levels since the Great Recession, even as wages rise and corporate profits are strong. AI is being used by both sides. Job seekers use ChatGPT to write and tailor resumes and cover letters. Employers are also using AI tools to screen, sort, and even interview candidates. Many applications do not reach real people, as automated systems and algorithmic filters decide who progresses. Many feel their resumes end up in a great void. When the human element is removed, such as a personal connection, it creates an increasingly impersonal situation. The takeaway is that more automation does not equal better hiring. It leaves applicants feeling stuck. While employers feel frustrated, unable to distinguish between talented and talentless candidates. #JobMarket #Careers #AIinHiring #ChatGPT #JobSearch #CareerAdvice #HumanCenteredWork https://lnkd.in/gGchUCt6

1

⚛️ Oak Ridge National Laboratory and Atomic Canyon signed a MoU to streamline licensing for nuclear power plants using #AI to speed up license application reviews. The agreement enables Atomic Canyon—a member of the NVIDIA Inception Program—to further develop its Neutron AI Platform, accelerated by NVIDIA, with enhanced cybersecurity features. ✨ Atomic Canyon developed its original FERMI AI models, which powers Neutron AI, using ORNL's Frontier supercomputer. ➡️ Read the announcement from Oak Ridge National Laboratory.

1

Video models don't just generate. They reason. And now we know exactly how. New research from UC Berkeley, UC San Diego, CMU and SenseTime just overturned a core assumption about how diffusion-based video models think. The old theory: reasoning unfolds frame by frame across a video. The reality: it happens across the denoising steps instead. Early steps explore multiple candidate solutions in parallel. Later steps narrow down to a single answer. When the authors disrupted specific denoising steps, performance collapsed. Disrupting individual frames barely mattered. Three emergent behaviors also showed up:  1. Working memory: Holding onto information even when objects disappear from view  2. Self-correction: Catching and reversing wrong answers mid-process  3. Perception before action: First figure out what and where, then figure out how The practical payoff: a simple, training-free method that ensembles outputs from multiple runs at the right internal layers. No fine-tuning. It outperformed Runway Gen-4 and Kling 2.6 on the VBVR-Bench. Really grateful to Ruisi Wang, Zhongang Cai, and all co-authors for this genuinely illuminating work. Link to Paper: https://lnkd.in/gUwR378V #GenerativeAI #VideoAI #DiffusionModels #AIResearch #DeepLearning

18

Metamates, here’s how to completely screw up today’s RSU vest: (I’ve done 4 of these) 1. Not calculating that your marginal tax rate is above the default 22% ↳ Bay Area FAANG salaries usually put you in 32%+ territory, not the RSU default 2. Don’t check your withholding rate ↳ It’s probably still set to that 22% while you actually owe 32%+ 3. Think holding = tax benefits ↳ Your RSUs are taxed as ordinary income whether you sell or hold 4. Hold for a year expecting long-term gains treatment ↳ Only the GAIN above vest price gets preferential treatment 5. Have zero plan for selling or deploying the money ↳ I’ve seen folks accumulate multiple 6-figure concentrated positions because they didn’t have a plan and didn’t take action What’s your biggest RSU lesson learned the hard way?

15

6 Comments

🧠 AI Rights Historical Wrongs Researchers at Stanford and Princeton have fine-tuned a large language model to uncover racially discriminatory housing clauses buried in millions of historical property deeds — accelerating a process that could have taken a decade by hand. 📍 The problem: Many property deeds in Northern California — especially in Santa Clara County — still contain illegal racial covenants from as far back as the 1850s, banning people of color from owning or living in certain homes. 🔎 The solution: The team used OCR to digitize 5.2 million deed pages, then fine-tuned Mistral-7B via LoRA on a hand-labeled dataset of discriminatory phrases. The model learned to identify harmful language in context, avoiding false positives like the surnames "Black" or "White." 📊 Results: Detected 24,500 lots with racial clauses (~25% of homes in 1950) Found that 10 developers were responsible for 1/3 of all discriminatory clauses Model reviewed all pages in 6 days at a cost of $258 Equivalent manual effort: 10 years, $1.4M 💥 This work doesn't just use LLMs to build the future — it uses them to understand and repair the past. The open-sourced model now enables other U.S. counties to follow suit. #AI #MachineLearning #LLM #ResponsibleAI #Stanford #Princeton #HistoryMeetsAI #TechForGood #EthicalAI #SocialImpact #PropertyTech https://lnkd.in/dtec-zbx

4

Recently, while working with a capstone project team of UCI students, I was reminded of an important point about industrial AI. Industrial AI today is often discussed through the lens of large language models such as ChatGPT, Gemini, Claude, or LLaMA. These technologies are impressive and have transformed how we interact with information. But in many industrial environments, real progress in AI depends on something different: 𝐝𝐨𝐦𝐚𝐢𝐧-𝐬𝐩𝐞𝐜𝐢𝐟𝐢𝐜 𝐦𝐚𝐜𝐡𝐢𝐧𝐞 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠. Industrial systems are fundamentally different from the text and images that general-purpose models are trained on. Engineering data appears in forms such as CAD geometry, sensor signals, time-series data, manufacturing processes, and physical simulations. These systems are governed by physics, constraints, and complex interactions between components. 𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐭𝐡𝐞𝐦 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐦𝐨𝐝𝐞𝐥𝐬 𝐝𝐞𝐬𝐢𝐠𝐧𝐞𝐝 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐨𝐟 𝐭𝐡𝐚𝐭 𝐝𝐚𝐭𝐚. For example, learning from CAD models requires geometric and topological reasoning. Predicting equipment behavior often relies on time-series analysis of sensor data. Manufacturing optimization may require graph-based representations of processes and assemblies. In these contexts, domain knowledge is not optional, it is essential. What we are beginning to see is a new layer of AI emerging in industry: models trained specifically on engineering and operational data. These models combine machine learning with physics, system constraints, and domain expertise. Rather than replacing engineering tools, they augment them and helping engineers analyze complex systems, detect patterns across large datasets, and explore design spaces more effectively. Large language models will certainly play a role, particularly as interfaces that help orchestrate workflows or make complex tools easier to access. But the real breakthroughs in industrial AI will likely come from deep integration between machine learning and domain-specific knowledge. In other words, the future of industrial AI will not be built by generic models alone. It will be built by 𝐜𝐨𝐦𝐛𝐢𝐧𝐢𝐧𝐠 𝐀𝐈 𝐰𝐢𝐭𝐡 𝐭𝐡𝐞 𝐝𝐞𝐞𝐩 𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐨𝐟 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐰𝐢𝐭𝐡𝐢𝐧 𝐞𝐚𝐜𝐡 𝐝𝐨𝐦𝐚𝐢𝐧. #ArtificialIntelligence #IndustrialAI #MachineLearning #DomainSpecificAI #EngineeringAI #DigitalTwin #Manufacturing #CAD #Industry40 #Engineering

3

february was insane for bay area AI funding: - waymo: $16B (yes, billion) - cerebras: $1B - elevenlabs: $500M - bedrock robotics: $270M that's $17.8B deployed in like 3 weeks. what this means: capital is flooding into production AI: stuff that ships and generates revenue (not research projects). autonomous vehicles, AI hardware, enterprise voice AI, construction robotics. all getting massive checks. I'm loving to connect with AI builders. Let's catch on 3/17. Link in the comments

26

2 Comments

I've been wanting to better document my projects, so I'm starting by showcasing what we accomplished in CSCE 633 Machine Learning at Texas A&M over the past semester! I implemented foundational ML algorithms from scratch and progressed to state-of-the-art deep learning architectures. Here's what I built: Classical ML Foundations: Implemented linear and logistic regression with gradient descent optimization Built decision trees from scratch using Gini impurity for classification Applied XGBoost with L1 regularization and bootstrap validation to handle imbalanced datasets Achieved 96% AUROC on binary classification tasks Support Vector Machines: Derived optimal separating hyperplanes by hand and implemented maximal margin classifiers Explored kernel methods (linear, RBF, polynomial) for non-linear decision boundaries Solved the classic XOR problem using feature transformation to higher dimensions Applied SVMs to graduate admissions prediction with 89% test accuracy Deep Learning - Computer Vision: Built a 5-layer CNN from scratch for scene classification on the SUN397 dataset Implemented batch normalization, dropout regularization, and data augmentation Achieved 78% accuracy classifying airport terminals, bedrooms, dining rooms, and living rooms Manually calculated convolution operations, stride effects, and max pooling to understand CNN mechanics Deep Learning - Natural Language Processing: Implemented bidirectional LSTMs and Transformer encoders for sentiment analysis Compared architectures on 25K IMDB movie reviews, both achieving 85%+ accuracy Built custom tokenization pipelines with vocabulary management and sequence padding Leveraged GPU acceleration (Google Colab) to reduce training time from 40+ hours to 30 minutes Key Technical Skills Developed: Python • PyTorch • scikit-learn • NumPy • Pandas • XGBoost • Data preprocessing • Hyperparameter tuning • K-fold cross-validation • Feature engineering • GPU computing Real-World Applications: These techniques directly apply to my target roles in perception engineering and computer vision for autonomous systems in defense - from classical ML for sensor fusion to CNNs for object detection and Transformers for multimodal AI. Check out my coursework repository: https://lnkd.in/gmFV8wip #MachineLearning #DeepLearning #AI #ComputerVision #NLP #TexasAM #GraduateSchool

2

Verification Funding Opportunity for CleanTech Companies: Through the SoCal OASIS® Initiative at University of California, Riverside, high-potential clean technology startups can receive up to $30,000 in support to conduct third-party verification or pilot testing at UCR facilities. This program is designed to help companies generate the performance data, validation, and technical support often needed to move from prototype to market readiness. Selected companies may receive access to UCR facilities and testing infrastructure, collaboration with technical experts, and greater visibility within the Inland Southern California innovation ecosystem. If your company is seeking data validation, performance testing, or third-party verification to unlock customer, investor, or partner confidence, we encourage you to apply. Learn more and apply: https://lnkd.in/gGun45wp

21

1 Comment

DeepSeek AI Signals a New Direction in Efficient Model Design : Latest Paper Suggests a Rethink of Large-Scale Training A newly published paper on arXiv proposes structural changes to neural networks that could meaningfully improve training stability and cost efficiency and it feels like a familiar pattern. The paper introduces mHC, a technique designed to stabilize large-scale training with minimal additional compute overhead. Tested on 3B, 9B, and 27B parameter models, mHC delivered stronger benchmark performance, particularly on reasoning-heavy tasks. Notably, CEO Liang Wenfeng co-authored and personally uploaded the paper, reinforcing how hands-on DeepSeek leadership remains with core research. If this sounds familiar, it should. Similar research breadcrumbs appeared ahead of R1 and V3, both of which preceded major releases. Why this matters Last year’s “DeepSeek moment” showed how R1 approached frontier-level performance at a fraction of the cost. This paper suggests they’re still unlocking efficiency gains. Combined with greater access to advanced AI chips, these architectural improvements point to something bigger. Looking ahead to 2026 Between cost-efficient training methods and accelerating hardware access, Chinese AI releases are set to become even more competitive—not just cheaper, but smarter and more stable at scale. This doesn’t look like a one-off paper. It looks like a preview. Read the paper here : https://lnkd.in/gR222q9P #AIResearch #DeepSeek #NeuralNetworks #LLMs #AIInfrastructure #MachineLearning #AITrends

7