Codenight Weekly

Edition: 043

Codenight is back with the latest!. This week's newsletter features amazing apps, web projects that defy expectations, and debates that spark innovation. Read it here.

#news #newsletters #subscribe #subscribe@CodeNight_Info

I am watching a lot of promo everywhere about the new TikTok like app developed for Ethiopians.

I am not sure about the level of investment made by the developer company (Infra Tech).
But, the main challenges are what called network effect, users join because their friends or creators from around the world are already there, and creators join because users are there, so it attract more users, It’s extremely difficult to break this loop When you think at large scale, like to get a million of users.

The second challenge is technical. TikTok is not just app, its AI feed requires advanced tracking, high expertise in machine learning, research, massive data processing, and powerful infrastructure to process all those videos with advanced compression and store for recommendation ,all of which are costly and complex. Trust, safety, moderation, and platform management also become major responsibilities.

So I hope they consider these realities beyond the hype they are creating. If they start with a small community first and scale gradually, we might see a successful social media platform emerge from Ethiopia.

I wish them the best of luck. I just hope they don’t end up delivering only a basic video player app.

If you are wondering to know about the trending conflict between Anthropic (Claude AI) and Pentagon (Department of War) here is the story in short


A few months ago, Anthropic signed a big $200 million deal to let the U.S. military use Claude on secret networks for defense tasks like intel analysis and planning.

Anthropic was actually the first big AI company to put its model on the military's super-secret classified networks. Everyone seemed happy.

Things went well, until the Pentagon demanded: "Remove all your limits. We need to use it for any lawful purpose, no exceptions."

Anthropic said no to two specific things:

- No mass surveillance of everyday American citizens inside the U.S.
- No fully autonomous weapons (AI that picks and kills targets without a human okay)

The Defense Secretary (Pete Hegseth) gave them a Friday deadline: drop those red lines or face consequences. Anthropic's CEO Dario Amodei refused, calling it a matter of safety and rights.

After the deadline passed:
- President Trump ordered all federal agencies to stop using Anthropic's AI (military gets 6 months to switch).
- The Pentagon labeled Anthropic a "supply chain risk" to national security (a tag usually for foreign threats like Chinese firms), blocking military contractors from working with them.
- Anthropic called it unfair and said they'll fight it in court.

OpenAI's response came fast: Hours later, they announced their own deal to put their models on classified DoD networks. CEO Sam Altman publicly said OpenAI shares the same red lines (no mass domestic surveillance, no fully autonomous offensive weapons). The Pentagon accepted their terms quickly.

DevMeetup V4 is coming — and we got a new home. 🚀

Check out the new site and let us know what you think.
More details about V4 dropping soon. Stay close.

https://new.devmeetup.et

Join our new channel https://t.me/devmeetupet

A list of countries that have developed LLM-scale models trained from scratch, from initial pre-training on large datasets to a final usable stage (publicly released, privately deployed, or government-backed). These models are trained independently and are not based on fine-tuning existing foreign models.

No LLM models built in Africa but there are some Initiatives
Source: wikipedia and internet

#AIModels #LLM

Just saw the news, OpenAI shut down the Sora app today, only six months after the massive hype of its launch.

The root cause of this massive failure seems to be copyright issues, the "ultimate killer" in this industry. After a wave of complaints from Hollywood and various rights holders, OpenAI tightened its content filters to avoid lawsuits, which completely frustrated users. Turns out the hype only lasted about a month before the reality set in.

#sora #openai

I completely agree with this speech.
great idea is worth no more than 5% of the final success. The remaining 95% comes down to two things, timing and execution.

Timing is everything. We’ve all seen incredibly simple ideas turn into million-dollar businesses not because the concept was revolutionary, but because the timing was perfect and/or the execution was relentless. Great execution can make an average idea shine, while poor execution can kill even the best one.

What fascinates me most is how universal ideas really are.

Almost every time a friend excitedly approaches me with “a new idea,” I’ve already thought about it, or know someone monetion similar idea. sometimes years earlier. If you’ve ever been at AASTU, you probably remember how many developers independently came up with the same concept …”building a campus delivery app” for cafes like K.K. It wasn’t one person’s stroke of genius it was something many of us had imagined.

Over time, I’ve come to believe that true “invention” is almost extinct in recent decades. What we call innovation today is usually just intelligent recombination connecting existing pieces in smarter, faster, or more convenient ways. I hope you know this qoute “If you want to reinvent the apple, you’d first have to create the entire universe.”

It’s also amusing how some people still try to hide their ideas, as if secrecy alone could protect them.

So the next time you get excited about a startup or product idea, remember this
Someone, somewhere, has probably already thought of it.

The real question you should ask yourself isn’t “Is this idea original?” but
“Why am I the one who can actually make this successful?”

That single shift in mindset changes everything.

@codeitlab

Dagmawi Bedilu is one of the thoughtful people I know. In this video, he discusses "Game Thinking," the startup ecosystem, and the education system in Ethiopia. It is definitely worth watching!

https://www.youtube.com/watch?v=GePpkI3H9X0

Anthropic accidentally leaked Claude Code's source code 😮

What happened:
On March 31, 2026, they released version 2.1.88 of the @anthropic-ai/claude-code npm package. It included a large source map file (cli.js.map ~60MB) that wasn't supposed to be there. This map pointed to a zip archive with the full unobfuscated TypeScript source.

What was exposed:
- Nearly 2,000 files
- Over 512,000 lines of code
- The entire CLI/agent architecture, tool-calling, streaming, telemetry, internal feature flags, and hints at unreleased features (like planning modes, undercover/anti-distillation tricks, etc.)

What was NOT exposed:
- Core Claude model weights
- Training data
- Any customer data or credentials

Anthropic confirmed it was human error in the build/packaging process (not a hack). They pulled the package and are adding safeguards. The code has already been mirrored widely on GitHub.

A wake-up call for release processes in fast-moving AI teams. The "harness" around the model is now public interesting times for agent devs!

Github is actively scanning and taking down recently created repositories of leaked source code.

#Claude #AI #Leak

Sometimes, I find myself pausing to consider how radically the era of software development has shifted. We have adopted AI into our workflows so deeply that the "old way" of writing code the manual, line-by-line construction of logic is starting to feel nearly impossible, or at least unnecessarily thing. It makes me wonder if we are experiencing a form of "learned helplessness" or a subtle "chatbot psychosis." It’s that creating sensation that you are incapable of solving a problem unless there is a prompt box waiting to catch your thoughts.

I often look back and ask mayself: How on earth did I build anything before 2023?

Fortunately, I have the evidence. I look at old applications I built years ago like AASTU Students App in 2021, now it help me as testament to my "past self." Back then, code was done by of granite. We wrote it line by line...we spent all day through Stack Overflow, copy-pasting snippets that failed 80% of the time, forcing us to actually understand the "why" behind the "how."

Yet, look at what we achieved. Before the era of LLM humans built the most complex systems on the planet we even built the AI itself. Millions of lines of hand-written code, integrated together by nothing but human persistence and logic. Those developers worldwide deserve a unique kind of honor...

Ultimately, I realize we aren't losing ourselves, we are simply doing what our species does best...which is adapting. The true beauty of human intelligence isn't found in the tools we use, It's in our uncanny ability to master a new reality and make it our own. We are not becoming weaker, we are becoming the orchestrators of a higher level of complexity.
@codeitlab

#AI #VibeCoding

https://claude.ai/settings/usage


claude is giving away some amount of reimbursement u might be eligible i got 20 usd u might get some check now.