OpenAI Admits AI Still Struggles With Most Coding Problems

OpenAI Admits AI Still Struggles With Most Coding Problems

Even the most advanced AI models can’t out-code humans. That’s what OpenAI researchers just confirmed, despite CEO Sam Altman’s bold claims that AI will outperform “low-level” coders by year-end.

In a new study, OpenAI’s own o1 reasoning model and GPT-4o, along with Anthropic’s Claude 3.5 Sonnet, were put to the test. The benchmark? SWE-Lancer, a set of 1,400 real-world coding tasks pulled from Upwork. The goal is to see how these AI models handle real freelance coding gigs.

Fast But Not Smart

The AI tackled two kinds of tasks -

They worked fast, way faster than humans. But speed isn’t everything. These models missed key issues, failed to find the root causes of bugs, and delivered half-baked fixes. Anyone who’s worked with AI knows this feeling—confident answers that fall apart on closer inspection.

AI Can’t Replace Real Engineers (Yet)

Even the best model, Claude 3.5 Sonnet, made more money than OpenAI’s models. But most of its solutions? Flat-out wrong. The researchers made it clear—these AIs need way more reliability before anyone should trust them with real coding work.

For now, humans still have the edge. AI can churn out quick fixes, but it doesn’t understand the big picture. It misses the deeper context, making mistakes no real software engineer would.

CEOs Are Moving Too Fast

Despite these clear limits, some CEOs are already firing human coders in favor of AI. The truth? AI isn’t ready to replace engineers. Not yet. Maybe not for a while. But that won’t stop companies from trying.

Apple Drops $500B On U.S. Expansion, Including AI Server Hub In Texas

Apple is making big moves. The tech giant just announced a massive $500 billion investment in the U.S. over the next four years. A new AI server factory in Houston, Texas, is part of the plan.

AI Factory on the Way

The 250,000-square-foot facility will build Apple Intelligence servers for iPhone, iPad, and Mac. It’s set to open in 2026. Apple says this factory will boost American manufacturing and support AI advancements.

20,000 New Jobs Incoming

Apple isn’t just building. It’s hiring 20,000 new employees across the U.S. The focus? R&D, silicon engineering, AI, and machine learning. The company is doubling down on its AI push.

Tim Cook: Betting Big on U.S. Tech

CEO Tim Cook says Apple is “bullish on American innovation”. He’s calling this investment a commitment to the future.

Trump, Tariffs, and U.S. Pressure

Apple’s decision comes after Tim Cook met with Donald Trump last week. The former president has been pressuring Apple to shift production away from China. Just this month, Trump slapped an extra 10% tariff on Chinese imports, on top of existing ones. Apple mostly assembles its products in China, so this shift matters.

More Than Just a Factory!

Apple’s $500 billion investment covers:

The company also announced it’s doubling its U.S. Advanced Manufacturing Fund from $5B to $10B.

Apple’s Tax Bill? Huge

Apple flexed its taxpayer status, saying it’s paid $75B in U.S. taxes over the last five years. Just in 2024? $19B. Apple is making a big bet on American tech. With AI heating up, this investment is just the beginning.

Next-Gen Weather Tech? NVIDIA Earth-2 Brings AI-Powered Super-Resolution

Forecasters need to know exactly where extreme weather will hit. That’s the key to better preparation. Enter NVIDIA CorrDiff—a game-changing AI weather model. It delivers ultra-detailed, kilometer-scale forecasts for wind, temperature, and rain. It’s part of NVIDIA Earth-2, a platform designed to simulate climate and weather conditions.

Big news. The science behind CorrDiff just landed in Communications Earth and Environment, a journal in the Nature portfolio. Even better, it’s now an NVIDIA NIM microservice. Weather tech companies, researchers, and agencies are already using it to boost forecasting power.

AI That Sees the Storm Before It Hits

Extreme weather is getting worse. Faster, more precise predictions save lives and protect communities. CorrDiff helps with everything—risk assessment, evacuations, disaster management, and climate-resilient infrastructure.

Agencies and startups worldwide are using CorrDiff and other Earth-2 tools to improve forecasts for extreme weather, renewable energy, and farming.

How Does It Work?

CorrDiff supercharges weather models using generative AI. It upgrades rough forecasts from 25 kilometers to just 2 kilometers. Think of it like AI that sharpens blurry images—but for weather.

Even better, it predicts new weather details missing from the input data, like radar reflectivity, which shows where and how hard it’s raining.

CorrDiff was trained on numerical weather models, generating 12x sharper forecasts. The first version—built for Taiwan—was created with help from Taiwan’s Central Weather Administration.

NVIDIA didn’t stop there. The latest CorrDiff microservice now covers the entire U.S. and is packed with real-world data on hurricanes, floods, tornadoes, and winter storms—all the big ones.

The U.S. version is 500x faster and 10,000x more energy-efficient than traditional weather forecasting on CPUs.

What’s Next?

The research team keeps pushing boundaries. New AI diffusion models are on the way. The goal? Better small-scale details and more accurate predictions for extreme weather events.

CorrDiff could even predict urban downwash—when powerful winds whip through city streets, damaging buildings and endangering pedestrians.

Global Adoption

Weather agencies worldwide are adopting CorrDiff. It’s helping regional forecasters, disaster response teams, and renewable energy planners get ahead of extreme weather.

In Taiwan, the National Science and Technology Center for Disaster Reduction is already using CorrDiff for disaster alerts. The results? More precise warnings and a gigawatt-hour of energy saved, thanks to NVIDIA AI’s efficiency.

CorrDiff forecasts are now integrated into Taiwan’s disaster monitoring system. This means better typhoon preparation, stronger forecasting, and faster response. With AI like this, weather forecasting is becoming faster, smarter, and more efficient than ever.

Indegene Just Launched Cortex; The Future Of AI In Life Sciences

Big changes are coming to healthcare in India. Indegene just dropped something massive; Cortex, a Generative AI (GenAI) platform built just for life sciences. This isn't some generic AI tool. It’s built for pharma and healthcare and designed to speed up innovation and boost efficiency across the industry.

Cortex: AI Built for Life Sciences

Cortex isn’t just another AI platform. It’s a specialist knowledge engine with multi-agent orchestration that helps scale GenAI safely. Indegene’s 25+ years of expertise power the system, making it smarter, faster, and tailored for industry challenges.

It’s tackling the biggest headaches in life sciences—content supply chain issues, medical writing, and slow Medical-Legal-Regulatory (MLR) reviews.

Why It Stands Out

Manish Gupta, Indegene’s CEO, says life sciences leaders believe in AI but want enterprise-grade security, governance, and scalability. Cortex delivers all that and more.

One big win? Multi-agent orchestration. This means seamless integration across entire enterprises. Plus, Cortex is flexible. It works with any preferred Large Language Model (LLM) and evolves with tech advancements. That means it keeps getting smarter over time.

AI That Adapts to You

Life sciences is changing fast, and leaders need AI that keeps up. Cortex does just that.

Tarun Mathur, Indegene’s CTO, explains it best—this isn’t just an AI tool. It’s a platform that lets organizations build their own AI IP while using Indegene’s expertise. That means more customization, smarter AI models, and better results.

Transforming Indian Healthcare

India’s healthcare industry is evolving, and Cortex is stepping up. This AI streamlines workflows, automates compliance, and enhances patient-provider experiences. It’s future-proofing the industry—making healthcare more personalized, efficient, and scalable.

The impact is massive.

HFS Research, a global business consultancy, named Indegene a Horizon 3 OneEcosystem Exponential Leader for its AI leadership in life sciences. Everest Group also recognized Indegene as a GenAI Front-Runner for its ability to scale AI, build partnerships, and deliver real results.

What’s Next?

As India’s digital health revolution gains speed, AI-powered platforms like Cortex will drive the future. Less paperwork. Faster approvals. Better decisions.

For healthcare leaders looking to modernize, Cortex isn’t just an AI platform—it’s the key to staying ahead.

Open-Source AI Gets Its Own Foundation to Ensure Ethical Innovation

The Open-Source AI Foundation is here. Its mission? More transparency. More accountability. Especially for AI used in civilian government agencies. The timing is no accident. DeepSeek just announced it’s open-sourcing parts of its AI models.

Transparency in AI? It’s About Time

Andrew Stiefel, Senior Product Marketing Manager at Endor Labs, is all for it. He compares this push for AI transparency to the U.S. government’s 2021 Executive Order on Cybersecurity. That order forced companies to provide a software bill of materials (SBOM) for products sold to the government. The goal? Track vulnerabilities. Improve security.

Applying the same logic to AI makes sense. Stiefel explains, "It’s not just about transparency for citizens and government employees. It’s also about security. Knowing what datasets, training methods, and model weights were used helps catch risks early."

DeepSeek's Open-Source Play—A Big Move

DeepSeek already made its models open-source, but now it's taking things further. Stiefel calls it a win for transparency and security.

"This step gives us a peek inside their hosted services," he says. "Now, we’ll see how they fine-tune and run their models in production. That means the community can audit them for security risks. It also makes it easier for people to run their own versions of DeepSeek."

Security matters. DeepSeek had a bit of a slip-up with unsecured Clickhouse databases. Stiefel thinks this move could set a precedent for other AI companies to be more open about their operations.

What Does "Open" Even Mean?

Julien Sobrier, Senior Product Manager at Endor Labs, wants clarity. "Open AI models are more than just code," he says. "They include training sets, weights, testing programs—everything. To be truly open, the whole chain needs to be available."

Right now, the definition is blurry. OpenAI, Meta, and others all have different takes on what 'open' really means. Sobrier warns about "open-washing"—companies pretending to be open while keeping key pieces locked down.

Open-Source AI With a Catch

Sobrier also sees a shift in open-source projects. Some companies offer paid versions of open-source software without contributing back. AI companies might follow.

"Meta and others could make their models 'more open,'" he says, "but still block competitors from using them." That’s a clever business move, but it shrinks the open-source spirit.

AI Models: The Next Big Risk?

Both Stiefel and Sobrier stress one thing—open-source AI needs proper risk management.

Stiefel points out, "Open-source AI models aren’t automatically riskier than closed ones. But they do need constant monitoring."

Sobrier takes it further. "AI models are dependencies, just like open-source software. Companies need to check if these models are legally safe to use. They also need to verify that training datasets weren’t poisoned or contained sensitive data."

The community must step up. AI needs best practices—ways to rate security, quality, operational risks, and openness. Without it, the risks only get bigger. The future of open AI? More transparency. More scrutiny. More accountability. And hopefully, less open-washing.

DeepSeek To Open-Source 5 Code Repositories For ‘Full Transparency’

DeepSeek just made a bold move. The Chinese AI start-up is doubling down on open-source technology. It’s putting five code repositories out in the open for anyone to use. This comes as the AI race between China and the U.S. heats up.

The Hangzhou-based company announced the news on X. Starting next week, its tiny team working on artificial general intelligence will share their progress. "No ivory towers. Just pure garage-energy and community-driven innovation," DeepSeek wrote.

Why This Matters

A code repository holds software. Developers can see, tweak, and contribute. Most companies guard these assets like treasure. But DeepSeek is flipping the script. Its V3 and R1 models are fully open-source. That means free to use and free to modify.

And that decision is paying off big.

DeepSeek’s low-cost, high-performance large language models (LLMs) sent shockwaves through the tech world. Investors scrambled. AI stocks soared. Since then, DeepSeek has kept quiet—until now.

The Research Continues

Earlier this week, DeepSeek dropped a study with its CEO, Liang Wenfeng, as one of 15 co-authors. It explores "native sparse attention"—a way to make LLMs process big data more efficiently.

This move shows that DeepSeek is serious about open research and development. Unlike OpenAI, Google, or Anthropic, which keep their best models under lock and key, DeepSeek is going all-in on transparency.

The AI Community Reacts

DeepSeek’s post on X has thousands of likes. On Reddit, AI enthusiasts called the company a "gift to humanity."

And DeepSeek isn’t stopping. It quietly updated its business registry this week. Analysts think this might mean LLM monetization is next. Big changes ahead. But one thing’s clear—DeepSeek is playing the open-source game at full speed.

Want to get Towards AGIfied?

Stay in touch with us by subscribing to our newsletter to receive weekly pieces and updates.

Please rate our newsletter below. Your feedback matters to us.

Thank you for reading

-Shen Pandi & Team