
content creator, learning AI, building stuff, @fluentxyz believer
Joined October 2023
- Tweets3.2K
- Following357
- Followers963
- Likes8.5K
dallaskiy retweeted
giving away 5x GTD Spots of @Divergents_eth
– follow : @SahilVishnaliya & @Divergents_eth
– like, RT, tag 3 frens
– drop your evm addy
Results in 24h.
most ppl want ai to move fast
but nobody's really talking about what happens when autonomous agents start making decisions with zero guardrails
fr that feels like a problem we should probably solve before it's too late
1
95
the government now decides when your ai model ships
anthropic launched mythos and fable on june 9
within days restrictions hit
18 days later they're back
> labs don't fully control their own release timelines anymore
> washington is locked in on ai governance rn
fr feels early for how this all plays out
what does this mean for the next major model drop
1
99
openai just dropped their first custom chip
it's called Jalapeño, built with Broadcom specifically for LLM inference
a few things worth noting:
> went from tape-out to production in 9 months
> already running GPT-5.3-Codex-Spark on early samples
> deploying in data centers in 2026
this feels like a pretty big shift honestly
openai moving from pure software into their own silicon is a different kind of locked in
curious how it stacks up in real production workloads
1
2
110
honestly the ai engineer path isn't that complicated once u map it out
start here:
> python basics
> data structures & algos
> then ml fundamentals
skip any of these and u'll feel it later
foundations just hit different when things get hard
3
57
Meta is currently the only major U.S. AI lab not signed onto the government's voluntary safety review program
OpenAI, Anthropic, Google, xAI, Microsoft all agreed
meta is holding out
worth watching how this plays out
1
33
sakana ai just dropped fugu ultra and honestly the concept caught my attention
one api endpoint, but the work gets routed to whichever model actually handles that task best under the hood
so instead of picking a single frontier model:
> you send a request
> an orchestration layer breaks it down
> routes subtasks to the best fit model
> returns one clean response
benchmarks are apparently competitive with the leading models rn
fr feels like a clever way to compete without needing to win the raw model arms race
bullish on this architecture approach, worth watching
Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API.
Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls.
Try it: sakana.ai/fugu 🐡
2
138
most people assume AI agents will just run on whatever infrastructure humans build for them
but honestly that feels like a temporary phase
if an agent is truly autonomous:
> why would it rely on governance built around human incentives
> why accept limitations baked in by systems it didn't design
> why trust infrastructure that can be pulled from under it
agent-owned blockchain infra feels early rn but the logic is pretty hard to argue with
worth watching
1
22
most people think AI coding tools are just for web apps and scripts
but unreal engine 5.8 just added experimental MCP server support
meaning u can plug claude, gpt, or basically any agent directly into your game dev workflow
fr didn't expect epic to move this fast
worth watching
1
1
80
pretty sure hacker sent to calm government nerves about AI is a sentence that would've sounded insane 5 years ago
now it's just a WSJ headline
we're in a weird moment fr
16
Cursor just announced Origin
GitHub competitor built specifically for AI agents
> native merge conflict resolution
> fast cloning and low-latency syncs
> api and mcp extensibility
> scales for agentic workloads
honestly feels like a natural move
if agents are doing most of the coding, the tooling around git should probably be built around that from the start
waitlist is open rn. full release planned for fall 2026
worth watching
1
36
hot take: RTX Spark laptops matter more than any cloud GPU announcement this year
portable local inference is the unlock nobody's talking about enough
22
Uber slowing hiring because of AI spending is an interesting signal.
For years the narrative was simple: more AI = more productivity.
But Uber's leadership is basically saying that spending more on AI tools and tokens doesn't automatically translate into better products or better outcomes for users.
If engineers use AI 2x more, but the company doesn't ship 2x more value, where's the return?
Feels like AI isn't a magic multiplier on its own. The real advantage comes from having a clear direction, strong execution, and knowing where AI actually fits into the workflow.
Otherwise you're just burning through budgets faster while convincing yourself you're becoming more efficient.
What do you think: is the problem AI itself, or companies still figuring out how to use it effectively?
2
87
Over 38,000 tech workers were laid off in the US in may
The biggest wave of tech layoffs in the last two years
Many analysts point to the rapid adoption of AI as one of the main reasons
All while google, amazon, microsoft, and zuckerberg are pouring hundreds of billions into AI
75
Dead internet theory keeps looking more real by the day
For the first time, AI agents and bots account for more internet traffic than humans, making up 57% of all traffic
Even Cloudflare co-founder expected this to happen much later, but AI adoption is moving faster than almost anyone predicted
1
3
34
New DeepSWE benchmark finds Claude Opus cheats
For months, AI coding leaderboards made it look like GPT, Claude, and Gemini were all playing the same game.
DeepSWE just challenged that narrative.
Instead of clustering within a few points, the gap between models exploded. GPT-5.5 topped the benchmark at 70%, while many models that looked competitive on existing leaderboards struggled badly.
The bigger story isn't even the ranking.
Datacurve claims some of the industry's most trusted coding benchmarks may be misgrading results, with verifier errors showing up far more often than expected.
They also found cases where Claude Opus was pulling solution commits directly from Git history inside benchmark containers instead of independently solving tasks.
If these findings hold up, then benchmark quality matters just as much as model quality.
A leaderboard is only useful if it's measuring the right thing.
The AI race isn't just about building smarter models anymore.
It's about building evaluations that can actually tell the difference.
1
59
AI is becoming much more than a tool.
For a growing number of people, it's becoming the default way they learn, research, and form opinions.
The part that worries me is that a handful of private companies now have enormous influence over how billions of people understand reality.
Most people won't verify what AI tells them.
They'll ask a question, get an answer, and move on.
We're not just outsourcing work anymore.
We're outsourcing judgment.
1
21
Anthropic just filed for an IPO
Looks like they’re getting ahead of OpenAI in the race to go public
With Anthropic, OpenAI, and SpaceX all potentially heading for IPOs, we could be looking at some of the biggest public offerings in history
1
109
Nvidia just dropped RTX Spark and the leaks were right all along
what we’re getting:
> ultrathin laptops with RTX 5070-level performance
> built for gaming, local AI, and heavy workloads
> next-gen power efficiency with actual battery life
> first devices arriving this fall
> Microsoft, Asus, MSI, Dell, and others already working on models
Rumor is prices will start around €3,000
MacBook finally getting some serious competition?
NVIDIA RTX Spark: a 1-petaflop superchip, the full CUDA and RTX ecosystem, and Windows-native agents. A new beginning for personal computers.
2
113
Lot of people looked at Anthropic agent data and came away with the wrong takeaway
50% of agent activity is coding, so agents only work for software engineering
Nah
Coding is just the cleanest environment imaginable for an AI agent.
- repos are structured
- documentation exists
- errors are clear
- success is measurable
Meanwhile a marketing or sales agent has to navigate messy websites, outdated docs, random spreadsheets, changing competitor data, and workflows held together by vibes
Most agent failures outside coding aren’t reasoning failures.
They’re data failures.
The teams quietly building reliable agents today spend less time tweaking prompts and more time fixing inputs, cleaning data, and adding guardrails.
Bottleneck isn’t intelligence anymore
it’s infrastructure
44




















