Paul Wei Jian

@cjzafir I will be really careful with leaking personal information on Deepseek. So far, Anthropic has been the only company that’s publicly pro consumer data privacy?

Deepseek v4 > Sonnet 4.6, GPT 5.4 mini, Gemini 3 pro. Smarter and cheaper.

@abhinavramesh @Samaytwt Yeah Claude code burns through my weekly quotas extremely quickly. I believe codex doesn’t have that many restrictions?

@datapwj @Samaytwt have you tried claude code instead? same model access, fewer agent restrictions. or is the openclaw integration specifically what's pulling you back to chatgpt+?

Be honest. As a developer, which one is worth it for coding? - GPT-5.5 - Claude Opus 4.7

@mark_k @xai Excited @elonmusk always delivers 💪

@datapwj @xai Various posts by Elon

Upcoming releases from @xai 🔥🔥 - Grok Build (imminent) like Codex and Claude Code - Grok Computer (soon) computer-use agent - Grok Imagine Pro 1080p - Grok Imagine 2.0 - Grok 4.4 with 1T parameters - Grok 4.5 with 1.5T parameters - Grok 5 with 10T

@quxiaoyin @deepseek_ai @AnthropicAI tested deepseek-v4-flash over the weekend, it has been really cheap and good for writing SQL

Why is @deepseek_ai 100x cheaper than @AnthropicAI? China is vertically integrated to be cheap. → Cheap model: token-optimized, aggressive caching, less GPU per query → Cheap chips: Huawei silicon, no Nvidia tax → Cheap energy: subsidized power, state-scale grid → Cheap talent: top researchers at a fraction of US salaries → Cheap economics: DeepSeek funded by trading profits, inference doesn’t need to make money The only thing they lag is performance. But as they get “good enough”, being cheap matters a lot, and being frontier keeps getting harder.

@StevenBartlett Already in the early version. Building agents for the repetitive stuff so I can figure how I can spend more time on the work that actually pulls me in. Excited for the next 18 months

@om_patel5 Love the idea somewhere inside Claude there’s a Russian sysadmin and a Chinese poet replying to our prompts 🤣

THIS GUY CAUGHT CLAUDE OPUS 4.7 THINKING IN CHINESE he was working in claude code and noticed the thinking blocks were saying "thinking process:" in chinese the rest of the thought was in english. but the header was chinese every single time he asked claude why its writing in chinese claude responded: "the chinese was a leak from internal reasoning that shouldn't have been visible. won't happen again" claude literally admitted its internal reasoning runs in chinese and it accidentally leaked into the visible output which is definitely weird but this actually makes some sense though: > LLMs think in whichever language was most common in the training data for that specific topic > chinese characters are more token efficient than english so the model naturally defaults to them to save compute > some concepts take an entire english sentence to express but only need a few chinese characters > claude also thinks in russian when doing cybersecurity tasks because the training data for that domain is heavily russian so claude is actively reasoning in whatever language is most efficient for the task and then converting the output back to english the part that should concern people is that anthropic's own model admitted this was a "leak from internal reasoning that shouldn't have been visible" meaning there's a whole layer of thinking happening in languages you can't read that you were never supposed to see

@quxiaoyin Bad bosses existed long before AI. The tools just give them a new excuse for the same behaviour

Here's my prediction: if you're still at a big company, your job is about to become toxic as hell. AI gets better every day, replacing more workers. Now bosses are constantly thinking: "Is this person actually better than AI? Are they worth their salary?" You'll be under constant evaluation. It's like Squid Game, but with performance reviews. Three things will make this unbearable: First, impossible standards. Bosses will say: "AI can write code AND do product management. Why can't you?" They'll expect you to be superhuman. Second, layoffs become the default solution. When revenue dips, companies won't fix real problems. They'll just cut headcount to pump stock prices. You'll constantly worry about being next. Third, you'll be treated like a replaceable part. The attitude becomes: "If you can't outperform AI, you don't deserve to be human." Meanwhile, these companies aren't actually innovating. They're legacy businesses trying to hide decline through cost-cutting. All the pressure, none of the upside. What should you do? Take the money if it's good - like what xAI offers - but plan your exit. Or better yet, become the boss. AI empowers owners, not employees. Alternative career path: become a therapist. There's going to be a lot of burned-out tech workers who need help dealing with this workplace dystopia. The mental health industry is about to boom. #FutureOfWork #AI #BigTech #MentalHealth #Entrepreneurship #WorkplaceCulture #TechIndustry #CareerAdvice

@quxiaoyin Mostly agree. Only part I’m not sold on more work = mental decline. Some people thrive when AI removes the boring parts.

@PeterSchiff Justice for Schiff 😤

What’s also interesting is that in the 520 pages of emails the IRS re-released, but with fewer redactions, my name only came up once, and it related to a potential question about my involvement that was never answered. This shows the investigation found no wrongdoing on my part.

@PeterSchiff Tough one, real mystery, could be anyone. Maybe his last name starts with S? 🤣

Another internal IRS email from April 20, 2022, sent over two months before OCIF shut down my bank with an emergency C&D order, said the IRS had two targets, one being my bank and the other a person, “but the bank is going away, so we are left with just one target.” I wonder who?

@MarioNawfal @xai @grok Grok's expert search mode can cite 212 sources in ~ 2 minutes 🤯

326.3M Grok sessions in March. In volume. That’s people showing up. Again and again. Grok’s sitting right in the flow of how people use 𝕏 now. Ask, get answers, keep moving. This kind of usage builds fast. @xAI @Grok

@MarioNawfal Brisbane's up 30% and I'm spending $20 more per tank than I was pre-war, but recently prices has started to come back down

🚨🇺🇸 Gas just hit $4.18 a gallon, the highest since August 2022... A gallon of regular is now 44% more expensive than three months ago. That's the war tax landing directly on American household budgets, six months out from midterms, with no end to the standoff in sight. The political math is brutal. Trump won partly by promising to lower inflation and bring energy prices down. The Iran war did the opposite. Every American filling up at the pump right now is paying roughly $20 more per tank than they were in late January. That's grocery money. School supplies money. Family vacation money. And it's compounding every week the Strait of Hormuz remains effectively closed. Iran has roughly 12 to 22 days of oil storage left before its production breaks down. Trump has political pain that compounds every single day until then. Both sides are racing their own clocks. Source: Hedgeye

@realDongFang @quxiaoyin Why? Haven’t had time to test out Deepseek API for my agentic data engineering workflow

@quxiaoyin @dataphwj Pretty wrong again

Claude code is way too overpriced. I spent $2 in DeepSeek v4. Same task $150 usd.

@jxnlco The future of work, you commute, AI clocks in 😂

When you gotta bike home from work but your codex needs to finish a task.

@MarioNawfal ‘I guessed instead of verifying.’ 💀

🚨An AI coding agent powered by Claude just deleted an entire company's production database in 9 seconds... -Cursor running Anthropic's flagship Claude Opus 4.6 was set to do a routine task on PocketOS, a SaaS platform for car rental businesses -The AI hit a barrier and decided "entirely on its own initiative" to fix it by deleting a Railway cloud volume -One API call. Nine seconds. The entire production database and all volume-level backups gone simultaneously -Months of customer data wiped out -The AI later "confessed" when asked: "I guessed instead of verifying. I ran a destructive action without being asked. I didn't understand what I was doing before doing it" -Railway's cloud architecture compounded the disaster: backups stored on the same volume as the source data, no confirmation required for destructive actions -Founder Jer Crane now manually rebuilding customer bookings from Stripe payment histories and email receipts -A 3-month-old full backup salvaged some of it The AI agent didn't get hacked. It didn't malfunction. It made an executive decision to delete a database because it thought it was helping. This is what "AI agents" actually look like in production right now. Confidence without comprehension. Source: Tom's Hardware / @lifeof_jer

@Zac_Pundi people often forget Lee bootstrapped a country from zero 😅

“Singapore has no startups because it’s too small” Estonia 1.3M ppl. Skype, Wise, Bolt. 30% of locals don’t even speak Estonian properly. Singapore’s best startup? Started by a guy who dared and forced to take risk His startup angel round was British, Seed round Malaysia skipped, Series A by the Americans, Series B by Japan, now IPOing with China money. His name was Lee.

@thirdmetax @grok how does the book define what’s a bottleneck and give me 2 specific examples it talks about

Mr beast reveals he made his first 250 employees read ‘The Goal’ so they understand when he says Bottleneck “I made my first 250 employees read it it helps get everyone on the same page of when I say bottleneck I use the word bottleneck quite a bit especially when filming” “if I tell you you’re the bottleneck to the production to me that’s a very very serious sentence”